Adding grounding dino (#26087)

* Fixed typo when converting weigths to GroundingDINO vision backbone * Final modifications on modeling * Removed unnecessary class * Fixed convert structure * Added image processing * make fixup partially completed * Now text_backbone_config has its own class * Modified convert script * Removed unnecessary config attribute * Added new function to generate sub sentence mask * Renamed parameters with gamma in the name as it's currently not allowed * Removed tokenization and image_processing scripts since we'll map from existing models * Fixed some issues with configuration * Just some modifications on conversion script * Other modifications * Copied deformable detr * First commit * Added bert to model * Bert validated * Created Text and Fusion layers for Encoder * Adapted Encoder layer * Fixed typos * Adjusted Encoder * Converted encoder to hf * Modified Decoder Layer * Modified main decoder class * Removed copy comments * Fixed forward from GroundingDINOModel and GroundingDINODecoder * Added all necessary layers, configurations and forward logic up to GroundingDINOModel * Added all layers to convertion * Fixed outputs for GroundingDINOModel and GroundingDINOForObjectDetection * Fixed mask input to encoders and fixed nn.MultiheadAttention batch first and attn output * Fixed forward from GroundingDINOTextEnhancerLayer * Fixed output bug with GroundingDINODeformableLayer * Fixed bugs that prevent GroundingDINOForObjectDetection to run forward method * Fixed attentions to be passed correctly * Passing temperature arg when creating Sine position embedding * Removed copy comments * Added temperature argument for position embedding * Fixed typo when converting weigths to GroundingDINO vision backbone * Final modifications on modeling * Removed unnecessary class * Fixed convert structure * Added image processing * make fixup partially completed * Now text_backbone_config has its own class * Modified convert script * Removed unnecessary config attribute * Added new function to generate sub sentence mask * Renamed parameters with gamma in the name as it's currently not allowed * Removed tokenization and image_processing scripts since we'll map from existing models * Fixed some issues with configuration * Just some modifications on conversion script * Other modifications * Fix style * Improve fixup * Improve conversion script * Improve conversion script * Add GroundingDINOProcessor * More improvements * Return token type ids * something * Fix more tests * More improvements * More cleanup * More improvements * Fixed tests, improved modeling and config * More improvements and fixing tests * Improved tests and modeling * Improved tests and added image processor * Improved tests inference * More improvements * More test improvements * Fixed last test * Improved docstrings and comments * Fix style * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Better naming * Better naming * Added Copied statement * Added Copied statement * Moved param init from GroundingDINOBiMultiHeadAttention * Better naming * Fixing clamp style * Better naming * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/models/grounding_dino/configuration_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/convert_grounding_dino_to_hf.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Improving conversion script * Improved config * Improved naming * Improved naming again * Improved grouding-dino.md * Moved grounding dino to multimodal * Update src/transformers/models/grounding_dino/convert_grounding_dino_to_hf.py Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> * Fixed docstrings and style * Fix docstrings * Remove timm attributes * Reorder imports * More improvements * Add Grounding DINO to pipeline * Remove model from check_repo * Added grounded post_process to GroundingDINOProcessor * Fixed style * Fixed GroundingDINOTextPrenetConfig docstrings * Aligned inputs.keys() when both image and text are passed with model_input_names * Added tests for GroundingDINOImageProcessor and GroundingDINOProcessor * Testing post_process_grounded_object_detection from GroundingDINOProcessor at test_inference_object_detection_head * Fixed order * Marked test with require_torch * Temporarily changed repo_id * More improvements * Fix style * Final improvements * Improve annotators * Fix style * Add is_torch_available * Remove type hints * vocab_tokens as one liner * Removed print statements * Renamed GroundingDINOTextPrenetConfig to GroundingDINOTextConfig * remove unnecessary comments * Removed unnecessary tests on conversion script * Renamed GroundingDINO to camel case GroundingDino * Fixed GroundingDinoProcessor docstrings * loading MSDA kernels in the modeling file * Fix copies * Replace nn.multiheadattention * Replace nn.multiheadattention * Fixed inputs for GroundingDinoMultiheadAttention & order of modules * Fixed processing to avoid messing with inputs * Added more tips for GroundingDino * Make style * Chaning name to align with SAM * Replace final nn.multiheadattention * Fix model tests * Update year, remove GenerationTesterMixin * Address comments * Address more comments * Rename TextPrenet to TextModel * Rename hidden_states * Address more comments * Address more comments * Address comment * Address more comments * Address merge * Address comment * Address comment * Address comment * Make style * Added layer norm eps to layer norms * Address more comments * More fixes * Fixed equivalence * Make fixup * Remove print statements * Address comments * Address comments * Address comments * Address comments * Address comments * Address comments * Add comment * Address comment * Remove overwriting of test * Fix bbox_embed * Improve decoder_bbox_embed_share * Simplify outputs * Updated post_process_grounded_object_detection * Renamed sources to feature_maps * Improved tests for Grounding Dino ImageProcessor and Processor * Fixed test requirements and imports * Fixed image_processing * Fixed processor tests * Fixed imports for image processing tests * Fix copies * Updated modeling * Fix style * Moved functions to correct position * Fixed copy issues * Update src/transformers/models/deformable_detr/modeling_deformable_detr.py Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> * Keeping consistency custom cuda kernels for MSDA * Make GroundingDinoProcessor logic clearer * Updated Grounding DINO checkpoints * Changed tests to correct structure * Updated gpu-cpu equivalence test * fix copies * Update src/transformers/models/grounding_dino/processing_grounding_dino.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/grounding_dino/processing_grounding_dino.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/grounding_dino/modeling_grounding_dino.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/grounding_dino/configuration_grounding_dino.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Fixed erros and style * Fix copies * Removed inheritance from PreTrainedModel from GroundingDinoTextModel * Fixed GroundingDinoTextModel * Fixed type of default backbone config * Fixed missing methods for GroundingDinoTextModel and Added timm support for GroundingDinoConvEncoder * Addressed comments * Addressed batched image processing tests * Addressed zero shot test comment * Addressed tip comment * Removed GroundingDinoTextModel from check_repo * Removed inplace masking * Addressed comments * Addressed comments * Addressed comments * Fix copies * Fixing timm test * Fixed batching equivalence test * Update docs/source/en/model_doc/grounding-dino.md Co-authored-by: Tianqi Xu <40522713+dandansamax@users.noreply.github.com> * Update docs/source/en/model_doc/grounding-dino.md Co-authored-by: Tianqi Xu <40522713+dandansamax@users.noreply.github.com> * Update docs/source/en/model_doc/grounding-dino.md Co-authored-by: Tianqi Xu <40522713+dandansamax@users.noreply.github.com> * Addressed more comments * Added a new comment * Reduced image size * Addressed more comments * Nits * Nits * Changed the way text_config is initialized * Update src/transformers/models/grounding_dino/processing_grounding_dino.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> --------- Co-authored-by: Niels <niels.rogge1@gmail.com> Co-authored-by: Rafael Padilla <31217453+rafaelpadilla@users.noreply.github.com> Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> Co-authored-by: Eduardo Pacheco <eduardo.pacheco@limehome.com> Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> Co-authored-by: Tianqi Xu <40522713+dandansamax@users.noreply.github.com>
2024-04-11 09:32:16 +02:00
parent a5e5c92aea
commit b752ad3019
36 changed files with 7411 additions and 20 deletions
--- a/tests/models/grounding_dino/init.py
+++ b/tests/models/grounding_dino/init.py
--- a/tests/models/grounding_dino/test_image_processing_grounding_dino.py
+++ b/tests/models/grounding_dino/test_image_processing_grounding_dino.py
@@ -0,0 +1,530 @@
+# coding=utf-8
+# Copyright 2024 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import json
+import pathlib
+import unittest
+
+from transformers.testing_utils import require_torch, require_vision, slow
+from transformers.utils import is_torch_available, is_vision_available
+
+from ...test_image_processing_common import AnnotationFormatTestMixin, ImageProcessingTestMixin, prepare_image_inputs
+
+
+if is_torch_available():
+    import torch
+
+    from transformers.models.grounding_dino.modeling_grounding_dino import GroundingDinoObjectDetectionOutput
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import GroundingDinoImageProcessor
+
+
+class GroundingDinoImageProcessingTester(unittest.TestCase):
+    def __init__(
+        self,
+        parent,
+        batch_size=7,
+        num_channels=3,
+        min_resolution=30,
+        max_resolution=400,
+        do_resize=True,
+        size=None,
+        do_normalize=True,
+        image_mean=[0.5, 0.5, 0.5],
+        image_std=[0.5, 0.5, 0.5],
+        do_rescale=True,
+        rescale_factor=1 / 255,
+        do_pad=True,
+    ):
+        # by setting size["longest_edge"] > max_resolution we're effectively not testing this :p
+        size = size if size is not None else {"shortest_edge": 18, "longest_edge": 1333}
+        self.parent = parent
+        self.batch_size = batch_size
+        self.num_channels = num_channels
+        self.min_resolution = min_resolution
+        self.max_resolution = max_resolution
+        self.do_resize = do_resize
+        self.size = size
+        self.do_normalize = do_normalize
+        self.image_mean = image_mean
+        self.image_std = image_std
+        self.do_rescale = do_rescale
+        self.rescale_factor = rescale_factor
+        self.do_pad = do_pad
+        self.num_queries = 5
+        self.embed_dim = 5
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTester.prepare_image_processor_dict with DeformableDetr->GroundingDino
+    def prepare_image_processor_dict(self):
+        return {
+            "do_resize": self.do_resize,
+            "size": self.size,
+            "do_normalize": self.do_normalize,
+            "image_mean": self.image_mean,
+            "image_std": self.image_std,
+            "do_rescale": self.do_rescale,
+            "rescale_factor": self.rescale_factor,
+            "do_pad": self.do_pad,
+        }
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTester.get_expected_values with DeformableDetr->GroundingDino
+    def get_expected_values(self, image_inputs, batched=False):
+        """
+        This function computes the expected height and width when providing images to GroundingDinoImageProcessor,
+        assuming do_resize is set to True with a scalar size.
+        """
+        if not batched:
+            image = image_inputs[0]
+            if isinstance(image, Image.Image):
+                w, h = image.size
+            else:
+                h, w = image.shape[1], image.shape[2]
+            if w < h:
+                expected_height = int(self.size["shortest_edge"] * h / w)
+                expected_width = self.size["shortest_edge"]
+            elif w > h:
+                expected_height = self.size["shortest_edge"]
+                expected_width = int(self.size["shortest_edge"] * w / h)
+            else:
+                expected_height = self.size["shortest_edge"]
+                expected_width = self.size["shortest_edge"]
+
+        else:
+            expected_values = []
+            for image in image_inputs:
+                expected_height, expected_width = self.get_expected_values([image])
+                expected_values.append((expected_height, expected_width))
+            expected_height = max(expected_values, key=lambda item: item[0])[0]
+            expected_width = max(expected_values, key=lambda item: item[1])[1]
+
+        return expected_height, expected_width
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTester.expected_output_image_shape with DeformableDetr->GroundingDino
+    def expected_output_image_shape(self, images):
+        height, width = self.get_expected_values(images, batched=True)
+        return self.num_channels, height, width
+
+    def get_fake_grounding_dino_output(self):
+        torch.manual_seed(42)
+        return GroundingDinoObjectDetectionOutput(
+            pred_boxes=torch.rand(self.batch_size, self.num_queries, 4),
+            logits=torch.rand(self.batch_size, self.num_queries, self.embed_dim),
+        )
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTester.prepare_image_inputs with DeformableDetr->GroundingDino
+    def prepare_image_inputs(self, equal_resolution=False, numpify=False, torchify=False):
+        return prepare_image_inputs(
+            batch_size=self.batch_size,
+            num_channels=self.num_channels,
+            min_resolution=self.min_resolution,
+            max_resolution=self.max_resolution,
+            equal_resolution=equal_resolution,
+            numpify=numpify,
+            torchify=torchify,
+        )
+
+
+@require_torch
+@require_vision
+class GroundingDinoImageProcessingTest(AnnotationFormatTestMixin, ImageProcessingTestMixin, unittest.TestCase):
+    image_processing_class = GroundingDinoImageProcessor if is_vision_available() else None
+
+    def setUp(self):
+        self.image_processor_tester = GroundingDinoImageProcessingTester(self)
+
+    @property
+    def image_processor_dict(self):
+        return self.image_processor_tester.prepare_image_processor_dict()
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTest.test_image_processor_properties with DeformableDetr->GroundingDino
+    def test_image_processor_properties(self):
+        image_processing = self.image_processing_class(**self.image_processor_dict)
+        self.assertTrue(hasattr(image_processing, "image_mean"))
+        self.assertTrue(hasattr(image_processing, "image_std"))
+        self.assertTrue(hasattr(image_processing, "do_normalize"))
+        self.assertTrue(hasattr(image_processing, "do_resize"))
+        self.assertTrue(hasattr(image_processing, "do_rescale"))
+        self.assertTrue(hasattr(image_processing, "do_pad"))
+        self.assertTrue(hasattr(image_processing, "size"))
+
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTest.test_image_processor_from_dict_with_kwargs with DeformableDetr->GroundingDino
+    def test_image_processor_from_dict_with_kwargs(self):
+        image_processor = self.image_processing_class.from_dict(self.image_processor_dict)
+        self.assertEqual(image_processor.size, {"shortest_edge": 18, "longest_edge": 1333})
+        self.assertEqual(image_processor.do_pad, True)
+
+        image_processor = self.image_processing_class.from_dict(
+            self.image_processor_dict, size=42, max_size=84, pad_and_return_pixel_mask=False
+        )
+        self.assertEqual(image_processor.size, {"shortest_edge": 42, "longest_edge": 84})
+        self.assertEqual(image_processor.do_pad, False)
+
+    def test_post_process_object_detection(self):
+        image_processor = self.image_processing_class(**self.image_processor_dict)
+        outputs = self.image_processor_tester.get_fake_grounding_dino_output()
+        results = image_processor.post_process_object_detection(outputs, threshold=0.0)
+
+        self.assertEqual(len(results), self.image_processor_tester.batch_size)
+        self.assertEqual(list(results[0].keys()), ["scores", "labels", "boxes"])
+        self.assertEqual(results[0]["boxes"].shape, (self.image_processor_tester.num_queries, 4))
+        self.assertEqual(results[0]["scores"].shape, (self.image_processor_tester.num_queries,))
+
+        expected_scores = torch.tensor([0.7050, 0.7222, 0.7222, 0.6829, 0.7220])
+        self.assertTrue(torch.allclose(results[0]["scores"], expected_scores, atol=1e-4))
+
+        expected_box_slice = torch.tensor([0.6908, 0.4354, 1.0737, 1.3947])
+        self.assertTrue(torch.allclose(results[0]["boxes"][0], expected_box_slice, atol=1e-4))
+
+    @slow
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTest.test_call_pytorch_with_coco_detection_annotations with DeformableDetr->GroundingDino
+    def test_call_pytorch_with_coco_detection_annotations(self):
+        # prepare image and target
+        image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+        with open("./tests/fixtures/tests_samples/COCO/coco_annotations.txt", "r") as f:
+            target = json.loads(f.read())
+
+        target = {"image_id": 39769, "annotations": target}
+
+        # encode them
+        image_processing = GroundingDinoImageProcessor()
+        encoding = image_processing(images=image, annotations=target, return_tensors="pt")
+
+        # verify pixel values
+        expected_shape = torch.Size([1, 3, 800, 1066])
+        self.assertEqual(encoding["pixel_values"].shape, expected_shape)
+
+        expected_slice = torch.tensor([0.2796, 0.3138, 0.3481])
+        self.assertTrue(torch.allclose(encoding["pixel_values"][0, 0, 0, :3], expected_slice, atol=1e-4))
+
+        # verify area
+        expected_area = torch.tensor([5887.9600, 11250.2061, 489353.8438, 837122.7500, 147967.5156, 165732.3438])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["area"], expected_area))
+        # verify boxes
+        expected_boxes_shape = torch.Size([6, 4])
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, expected_boxes_shape)
+        expected_boxes_slice = torch.tensor([0.5503, 0.2765, 0.0604, 0.2215])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"][0], expected_boxes_slice, atol=1e-3))
+        # verify image_id
+        expected_image_id = torch.tensor([39769])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["image_id"], expected_image_id))
+        # verify is_crowd
+        expected_is_crowd = torch.tensor([0, 0, 0, 0, 0, 0])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["iscrowd"], expected_is_crowd))
+        # verify class_labels
+        expected_class_labels = torch.tensor([75, 75, 63, 65, 17, 17])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["class_labels"], expected_class_labels))
+        # verify orig_size
+        expected_orig_size = torch.tensor([480, 640])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["orig_size"], expected_orig_size))
+        # verify size
+        expected_size = torch.tensor([800, 1066])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["size"], expected_size))
+
+    @slow
+    # Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_batched_coco_detection_annotations with Detr->GroundingDino
+    def test_batched_coco_detection_annotations(self):
+        image_0 = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+        image_1 = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png").resize((800, 800))
+
+        with open("./tests/fixtures/tests_samples/COCO/coco_annotations.txt", "r") as f:
+            target = json.loads(f.read())
+
+        annotations_0 = {"image_id": 39769, "annotations": target}
+        annotations_1 = {"image_id": 39769, "annotations": target}
+
+        # Adjust the bounding boxes for the resized image
+        w_0, h_0 = image_0.size
+        w_1, h_1 = image_1.size
+        for i in range(len(annotations_1["annotations"])):
+            coords = annotations_1["annotations"][i]["bbox"]
+            new_bbox = [
+                coords[0] * w_1 / w_0,
+                coords[1] * h_1 / h_0,
+                coords[2] * w_1 / w_0,
+                coords[3] * h_1 / h_0,
+            ]
+            annotations_1["annotations"][i]["bbox"] = new_bbox
+
+        images = [image_0, image_1]
+        annotations = [annotations_0, annotations_1]
+
+        image_processing = GroundingDinoImageProcessor()
+        encoding = image_processing(
+            images=images,
+            annotations=annotations,
+            return_segmentation_masks=True,
+            return_tensors="pt",  # do_convert_annotations=True
+        )
+
+        # Check the pixel values have been padded
+        postprocessed_height, postprocessed_width = 800, 1066
+        expected_shape = torch.Size([2, 3, postprocessed_height, postprocessed_width])
+        self.assertEqual(encoding["pixel_values"].shape, expected_shape)
+
+        # Check the bounding boxes have been adjusted for padded images
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, torch.Size([6, 4]))
+        self.assertEqual(encoding["labels"][1]["boxes"].shape, torch.Size([6, 4]))
+        expected_boxes_0 = torch.tensor(
+            [
+                [0.6879, 0.4609, 0.0755, 0.3691],
+                [0.2118, 0.3359, 0.2601, 0.1566],
+                [0.5011, 0.5000, 0.9979, 1.0000],
+                [0.5010, 0.5020, 0.9979, 0.9959],
+                [0.3284, 0.5944, 0.5884, 0.8112],
+                [0.8394, 0.5445, 0.3213, 0.9110],
+            ]
+        )
+        expected_boxes_1 = torch.tensor(
+            [
+                [0.4130, 0.2765, 0.0453, 0.2215],
+                [0.1272, 0.2016, 0.1561, 0.0940],
+                [0.3757, 0.4933, 0.7488, 0.9865],
+                [0.3759, 0.5002, 0.7492, 0.9955],
+                [0.1971, 0.5456, 0.3532, 0.8646],
+                [0.5790, 0.4115, 0.3430, 0.7161],
+            ]
+        )
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"], expected_boxes_0, rtol=1e-3))
+        self.assertTrue(torch.allclose(encoding["labels"][1]["boxes"], expected_boxes_1, rtol=1e-3))
+
+        # Check the masks have also been padded
+        self.assertEqual(encoding["labels"][0]["masks"].shape, torch.Size([6, 800, 1066]))
+        self.assertEqual(encoding["labels"][1]["masks"].shape, torch.Size([6, 800, 1066]))
+
+        # Check if do_convert_annotations=False, then the annotations are not converted to centre_x, centre_y, width, height
+        # format and not in the range [0, 1]
+        encoding = image_processing(
+            images=images,
+            annotations=annotations,
+            return_segmentation_masks=True,
+            do_convert_annotations=False,
+            return_tensors="pt",
+        )
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, torch.Size([6, 4]))
+        self.assertEqual(encoding["labels"][1]["boxes"].shape, torch.Size([6, 4]))
+        # Convert to absolute coordinates
+        unnormalized_boxes_0 = torch.vstack(
+            [
+                expected_boxes_0[:, 0] * postprocessed_width,
+                expected_boxes_0[:, 1] * postprocessed_height,
+                expected_boxes_0[:, 2] * postprocessed_width,
+                expected_boxes_0[:, 3] * postprocessed_height,
+            ]
+        ).T
+        unnormalized_boxes_1 = torch.vstack(
+            [
+                expected_boxes_1[:, 0] * postprocessed_width,
+                expected_boxes_1[:, 1] * postprocessed_height,
+                expected_boxes_1[:, 2] * postprocessed_width,
+                expected_boxes_1[:, 3] * postprocessed_height,
+            ]
+        ).T
+        # Convert from centre_x, centre_y, width, height to x_min, y_min, x_max, y_max
+        expected_boxes_0 = torch.vstack(
+            [
+                unnormalized_boxes_0[:, 0] - unnormalized_boxes_0[:, 2] / 2,
+                unnormalized_boxes_0[:, 1] - unnormalized_boxes_0[:, 3] / 2,
+                unnormalized_boxes_0[:, 0] + unnormalized_boxes_0[:, 2] / 2,
+                unnormalized_boxes_0[:, 1] + unnormalized_boxes_0[:, 3] / 2,
+            ]
+        ).T
+        expected_boxes_1 = torch.vstack(
+            [
+                unnormalized_boxes_1[:, 0] - unnormalized_boxes_1[:, 2] / 2,
+                unnormalized_boxes_1[:, 1] - unnormalized_boxes_1[:, 3] / 2,
+                unnormalized_boxes_1[:, 0] + unnormalized_boxes_1[:, 2] / 2,
+                unnormalized_boxes_1[:, 1] + unnormalized_boxes_1[:, 3] / 2,
+            ]
+        ).T
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"], expected_boxes_0, rtol=1))
+        self.assertTrue(torch.allclose(encoding["labels"][1]["boxes"], expected_boxes_1, rtol=1))
+
+    @slow
+    # Copied from tests.models.deformable_detr.test_image_processing_deformable_detr.DeformableDetrImageProcessingTest.test_call_pytorch_with_coco_panoptic_annotations with DeformableDetr->GroundingDino
+    def test_call_pytorch_with_coco_panoptic_annotations(self):
+        # prepare image, target and masks_path
+        image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+        with open("./tests/fixtures/tests_samples/COCO/coco_panoptic_annotations.txt", "r") as f:
+            target = json.loads(f.read())
+
+        target = {"file_name": "000000039769.png", "image_id": 39769, "segments_info": target}
+
+        masks_path = pathlib.Path("./tests/fixtures/tests_samples/COCO/coco_panoptic")
+
+        # encode them
+        image_processing = GroundingDinoImageProcessor(format="coco_panoptic")
+        encoding = image_processing(images=image, annotations=target, masks_path=masks_path, return_tensors="pt")
+
+        # verify pixel values
+        expected_shape = torch.Size([1, 3, 800, 1066])
+        self.assertEqual(encoding["pixel_values"].shape, expected_shape)
+
+        expected_slice = torch.tensor([0.2796, 0.3138, 0.3481])
+        self.assertTrue(torch.allclose(encoding["pixel_values"][0, 0, 0, :3], expected_slice, atol=1e-4))
+
+        # verify area
+        expected_area = torch.tensor([147979.6875, 165527.0469, 484638.5938, 11292.9375, 5879.6562, 7634.1147])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["area"], expected_area))
+        # verify boxes
+        expected_boxes_shape = torch.Size([6, 4])
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, expected_boxes_shape)
+        expected_boxes_slice = torch.tensor([0.2625, 0.5437, 0.4688, 0.8625])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"][0], expected_boxes_slice, atol=1e-3))
+        # verify image_id
+        expected_image_id = torch.tensor([39769])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["image_id"], expected_image_id))
+        # verify is_crowd
+        expected_is_crowd = torch.tensor([0, 0, 0, 0, 0, 0])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["iscrowd"], expected_is_crowd))
+        # verify class_labels
+        expected_class_labels = torch.tensor([17, 17, 63, 75, 75, 93])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["class_labels"], expected_class_labels))
+        # verify masks
+        expected_masks_sum = 822873
+        self.assertEqual(encoding["labels"][0]["masks"].sum().item(), expected_masks_sum)
+        # verify orig_size
+        expected_orig_size = torch.tensor([480, 640])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["orig_size"], expected_orig_size))
+        # verify size
+        expected_size = torch.tensor([800, 1066])
+        self.assertTrue(torch.allclose(encoding["labels"][0]["size"], expected_size))
+
+    @slow
+    # Copied from tests.models.detr.test_image_processing_detr.DetrImageProcessingTest.test_batched_coco_panoptic_annotations with Detr->GroundingDino
+    def test_batched_coco_panoptic_annotations(self):
+        # prepare image, target and masks_path
+        image_0 = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+        image_1 = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png").resize((800, 800))
+
+        with open("./tests/fixtures/tests_samples/COCO/coco_panoptic_annotations.txt", "r") as f:
+            target = json.loads(f.read())
+
+        annotation_0 = {"file_name": "000000039769.png", "image_id": 39769, "segments_info": target}
+        annotation_1 = {"file_name": "000000039769.png", "image_id": 39769, "segments_info": target}
+
+        w_0, h_0 = image_0.size
+        w_1, h_1 = image_1.size
+        for i in range(len(annotation_1["segments_info"])):
+            coords = annotation_1["segments_info"][i]["bbox"]
+            new_bbox = [
+                coords[0] * w_1 / w_0,
+                coords[1] * h_1 / h_0,
+                coords[2] * w_1 / w_0,
+                coords[3] * h_1 / h_0,
+            ]
+            annotation_1["segments_info"][i]["bbox"] = new_bbox
+
+        masks_path = pathlib.Path("./tests/fixtures/tests_samples/COCO/coco_panoptic")
+
+        images = [image_0, image_1]
+        annotations = [annotation_0, annotation_1]
+
+        # encode them
+        image_processing = GroundingDinoImageProcessor(format="coco_panoptic")
+        encoding = image_processing(
+            images=images,
+            annotations=annotations,
+            masks_path=masks_path,
+            return_tensors="pt",
+            return_segmentation_masks=True,
+        )
+
+        # Check the pixel values have been padded
+        postprocessed_height, postprocessed_width = 800, 1066
+        expected_shape = torch.Size([2, 3, postprocessed_height, postprocessed_width])
+        self.assertEqual(encoding["pixel_values"].shape, expected_shape)
+
+        # Check the bounding boxes have been adjusted for padded images
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, torch.Size([6, 4]))
+        self.assertEqual(encoding["labels"][1]["boxes"].shape, torch.Size([6, 4]))
+        expected_boxes_0 = torch.tensor(
+            [
+                [0.2625, 0.5437, 0.4688, 0.8625],
+                [0.7719, 0.4104, 0.4531, 0.7125],
+                [0.5000, 0.4927, 0.9969, 0.9854],
+                [0.1688, 0.2000, 0.2063, 0.0917],
+                [0.5492, 0.2760, 0.0578, 0.2187],
+                [0.4992, 0.4990, 0.9984, 0.9979],
+            ]
+        )
+        expected_boxes_1 = torch.tensor(
+            [
+                [0.1576, 0.3262, 0.2814, 0.5175],
+                [0.4634, 0.2463, 0.2720, 0.4275],
+                [0.3002, 0.2956, 0.5985, 0.5913],
+                [0.1013, 0.1200, 0.1238, 0.0550],
+                [0.3297, 0.1656, 0.0347, 0.1312],
+                [0.2997, 0.2994, 0.5994, 0.5987],
+            ]
+        )
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"], expected_boxes_0, rtol=1e-3))
+        self.assertTrue(torch.allclose(encoding["labels"][1]["boxes"], expected_boxes_1, rtol=1e-3))
+
+        # Check the masks have also been padded
+        self.assertEqual(encoding["labels"][0]["masks"].shape, torch.Size([6, 800, 1066]))
+        self.assertEqual(encoding["labels"][1]["masks"].shape, torch.Size([6, 800, 1066]))
+
+        # Check if do_convert_annotations=False, then the annotations are not converted to centre_x, centre_y, width, height
+        # format and not in the range [0, 1]
+        encoding = image_processing(
+            images=images,
+            annotations=annotations,
+            masks_path=masks_path,
+            return_segmentation_masks=True,
+            do_convert_annotations=False,
+            return_tensors="pt",
+        )
+        self.assertEqual(encoding["labels"][0]["boxes"].shape, torch.Size([6, 4]))
+        self.assertEqual(encoding["labels"][1]["boxes"].shape, torch.Size([6, 4]))
+        # Convert to absolute coordinates
+        unnormalized_boxes_0 = torch.vstack(
+            [
+                expected_boxes_0[:, 0] * postprocessed_width,
+                expected_boxes_0[:, 1] * postprocessed_height,
+                expected_boxes_0[:, 2] * postprocessed_width,
+                expected_boxes_0[:, 3] * postprocessed_height,
+            ]
+        ).T
+        unnormalized_boxes_1 = torch.vstack(
+            [
+                expected_boxes_1[:, 0] * postprocessed_width,
+                expected_boxes_1[:, 1] * postprocessed_height,
+                expected_boxes_1[:, 2] * postprocessed_width,
+                expected_boxes_1[:, 3] * postprocessed_height,
+            ]
+        ).T
+        # Convert from centre_x, centre_y, width, height to x_min, y_min, x_max, y_max
+        expected_boxes_0 = torch.vstack(
+            [
+                unnormalized_boxes_0[:, 0] - unnormalized_boxes_0[:, 2] / 2,
+                unnormalized_boxes_0[:, 1] - unnormalized_boxes_0[:, 3] / 2,
+                unnormalized_boxes_0[:, 0] + unnormalized_boxes_0[:, 2] / 2,
+                unnormalized_boxes_0[:, 1] + unnormalized_boxes_0[:, 3] / 2,
+            ]
+        ).T
+        expected_boxes_1 = torch.vstack(
+            [
+                unnormalized_boxes_1[:, 0] - unnormalized_boxes_1[:, 2] / 2,
+                unnormalized_boxes_1[:, 1] - unnormalized_boxes_1[:, 3] / 2,
+                unnormalized_boxes_1[:, 0] + unnormalized_boxes_1[:, 2] / 2,
+                unnormalized_boxes_1[:, 1] + unnormalized_boxes_1[:, 3] / 2,
+            ]
+        ).T
+        self.assertTrue(torch.allclose(encoding["labels"][0]["boxes"], expected_boxes_0, rtol=1))
+        self.assertTrue(torch.allclose(encoding["labels"][1]["boxes"], expected_boxes_1, rtol=1))
--- a/tests/models/grounding_dino/test_modeling_grounding_dino.py
+++ b/tests/models/grounding_dino/test_modeling_grounding_dino.py
@@ -0,0 +1,689 @@
+# coding=utf-8
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Testing suite for the PyTorch Grounding DINO model. """
+
+import collections
+import inspect
+import math
+import re
+import unittest
+
+from transformers import (
+    GroundingDinoConfig,
+    SwinConfig,
+    is_torch_available,
+    is_vision_available,
+)
+from transformers.file_utils import cached_property
+from transformers.testing_utils import (
+    require_timm,
+    require_torch,
+    require_torch_gpu,
+    require_vision,
+    slow,
+    torch_device,
+)
+
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import GroundingDinoForObjectDetection, GroundingDinoModel
+    from transformers.pytorch_utils import id_tensor_storage
+
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import AutoProcessor
+
+
+class GroundingDinoModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=4,
+        is_training=True,
+        use_labels=True,
+        hidden_size=32,
+        num_hidden_layers=2,
+        num_attention_heads=4,
+        intermediate_size=4,
+        hidden_act="gelu",
+        hidden_dropout_prob=0.1,
+        attention_probs_dropout_prob=0.1,
+        num_queries=2,
+        num_channels=3,
+        image_size=98,
+        n_targets=8,
+        num_labels=3,
+        num_feature_levels=4,
+        encoder_n_points=2,
+        decoder_n_points=6,
+        max_text_len=7,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.is_training = is_training
+        self.use_labels = use_labels
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.hidden_act = hidden_act
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.num_queries = num_queries
+        self.num_channels = num_channels
+        self.image_size = image_size
+        self.n_targets = n_targets
+        self.num_labels = num_labels
+        self.num_feature_levels = num_feature_levels
+        self.encoder_n_points = encoder_n_points
+        self.decoder_n_points = decoder_n_points
+        self.max_text_len = max_text_len
+
+        # we also set the expected seq length for both encoder and decoder
+        self.encoder_seq_length_vision = (
+            math.ceil(self.image_size / 8) ** 2
+            + math.ceil(self.image_size / 16) ** 2
+            + math.ceil(self.image_size / 32) ** 2
+            + math.ceil(self.image_size / 64) ** 2
+        )
+
+        self.encoder_seq_length_text = self.max_text_len
+
+        self.decoder_seq_length = self.num_queries
+
+    def prepare_config_and_inputs(self):
+        pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
+        pixel_mask = torch.ones([self.batch_size, self.image_size, self.image_size], device=torch_device)
+
+        input_ids = ids_tensor([self.batch_size, self.max_text_len], self.num_labels)
+
+        labels = None
+        if self.use_labels:
+            # labels is a list of Dict (each Dict being the labels for a given example in the batch)
+            labels = []
+            for i in range(self.batch_size):
+                target = {}
+                target["class_labels"] = torch.randint(
+                    high=self.num_labels, size=(self.n_targets,), device=torch_device
+                )
+                target["boxes"] = torch.rand(self.n_targets, 4, device=torch_device)
+                target["masks"] = torch.rand(self.n_targets, self.image_size, self.image_size, device=torch_device)
+                labels.append(target)
+
+        config = self.get_config()
+        return config, pixel_values, pixel_mask, input_ids, labels
+
+    def get_config(self):
+        swin_config = SwinConfig(
+            window_size=7,
+            embed_dim=8,
+            depths=[1, 1, 1, 1],
+            num_heads=[1, 1, 1, 1],
+            image_size=self.image_size,
+            out_features=["stage2", "stage3", "stage4"],
+            out_indices=[2, 3, 4],
+        )
+        text_backbone = {
+            "hidden_size": 8,
+            "num_hidden_layers": 2,
+            "num_attention_heads": 2,
+            "intermediate_size": 8,
+            "max_position_embeddings": 8,
+            "model_type": "bert",
+        }
+        return GroundingDinoConfig(
+            d_model=self.hidden_size,
+            encoder_layers=self.num_hidden_layers,
+            decoder_layers=self.num_hidden_layers,
+            encoder_attention_heads=self.num_attention_heads,
+            decoder_attention_heads=self.num_attention_heads,
+            encoder_ffn_dim=self.intermediate_size,
+            decoder_ffn_dim=self.intermediate_size,
+            dropout=self.hidden_dropout_prob,
+            attention_dropout=self.attention_probs_dropout_prob,
+            num_queries=self.num_queries,
+            num_labels=self.num_labels,
+            num_feature_levels=self.num_feature_levels,
+            encoder_n_points=self.encoder_n_points,
+            decoder_n_points=self.decoder_n_points,
+            use_timm_backbone=False,
+            backbone_config=swin_config,
+            max_text_len=self.max_text_len,
+            text_config=text_backbone,
+        )
+
+    def prepare_config_and_inputs_for_common(self):
+        config, pixel_values, pixel_mask, input_ids, labels = self.prepare_config_and_inputs()
+        inputs_dict = {"pixel_values": pixel_values, "pixel_mask": pixel_mask, "input_ids": input_ids}
+        return config, inputs_dict
+
+    def create_and_check_model(self, config, pixel_values, pixel_mask, input_ids, labels):
+        model = GroundingDinoModel(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        result = model(pixel_values=pixel_values, pixel_mask=pixel_mask, input_ids=input_ids)
+
+        self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.num_queries, self.hidden_size))
+
+    def create_and_check_object_detection_head_model(self, config, pixel_values, pixel_mask, input_ids, labels):
+        model = GroundingDinoForObjectDetection(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        result = model(pixel_values=pixel_values, pixel_mask=pixel_mask, input_ids=input_ids)
+
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_queries, config.max_text_len))
+        self.parent.assertEqual(result.pred_boxes.shape, (self.batch_size, self.num_queries, 4))
+
+        result = model(pixel_values=pixel_values, pixel_mask=pixel_mask, input_ids=input_ids, labels=labels)
+
+        self.parent.assertEqual(result.loss.shape, ())
+        self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_queries, config.max_text_len))
+        self.parent.assertEqual(result.pred_boxes.shape, (self.batch_size, self.num_queries, 4))
+
+
+@require_torch
+class GroundingDinoModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
+    all_model_classes = (GroundingDinoModel, GroundingDinoForObjectDetection) if is_torch_available() else ()
+    is_encoder_decoder = True
+    test_torchscript = False
+    test_pruning = False
+    test_head_masking = False
+    test_missing_keys = False
+    pipeline_model_mapping = (
+        {"image-feature-extraction": GroundingDinoModel, "zero-shot-object-detection": GroundingDinoForObjectDetection}
+        if is_torch_available()
+        else {}
+    )
+
+    # special case for head models
+    def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
+        inputs_dict = super()._prepare_for_class(inputs_dict, model_class, return_labels=return_labels)
+
+        if return_labels:
+            if model_class.__name__ == "GroundingDinoForObjectDetection":
+                labels = []
+                for i in range(self.model_tester.batch_size):
+                    target = {}
+                    target["class_labels"] = torch.ones(
+                        size=(self.model_tester.n_targets,), device=torch_device, dtype=torch.long
+                    )
+                    target["boxes"] = torch.ones(
+                        self.model_tester.n_targets, 4, device=torch_device, dtype=torch.float
+                    )
+                    target["masks"] = torch.ones(
+                        self.model_tester.n_targets,
+                        self.model_tester.image_size,
+                        self.model_tester.image_size,
+                        device=torch_device,
+                        dtype=torch.float,
+                    )
+                    labels.append(target)
+                inputs_dict["labels"] = labels
+
+        return inputs_dict
+
+    def setUp(self):
+        self.model_tester = GroundingDinoModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=GroundingDinoConfig, has_text_modality=False)
+
+    def test_config(self):
+        # we don't test common_properties and arguments_init as these don't apply for Grounding DINO
+        self.config_tester.create_and_test_config_to_json_string()
+        self.config_tester.create_and_test_config_to_json_file()
+        self.config_tester.create_and_test_config_from_and_save_pretrained()
+        self.config_tester.create_and_test_config_with_num_labels()
+        self.config_tester.check_config_can_be_init_without_params()
+
+    def test_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_model(*config_and_inputs)
+
+    def test_object_detection_head_model(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_object_detection_head_model(*config_and_inputs)
+
+    @unittest.skip(reason="Grounding DINO does not use inputs_embeds")
+    def test_inputs_embeds(self):
+        pass
+
+    @unittest.skip(reason="Grounding DINO does not have a get_input_embeddings method")
+    def test_model_common_attributes(self):
+        pass
+
+    @unittest.skip(reason="Grounding DINO does not use token embeddings")
+    def test_resize_tokens_embeddings(self):
+        pass
+
+    @unittest.skip(reason="Feed forward chunking is not implemented")
+    def test_feed_forward_chunking(self):
+        pass
+
+    def test_attention_outputs(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        config.return_dict = True
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_attentions"] = True
+            inputs_dict["output_hidden_states"] = False
+            config.return_dict = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+            attentions = outputs.encoder_attentions[-1]
+            self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
+
+            # check that output_attentions also work using config
+            del inputs_dict["output_attentions"]
+            config.output_attentions = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+            attentions = outputs.encoder_attentions[-1]
+            self.assertEqual(len(attentions), self.model_tester.num_hidden_layers)
+
+            self.assertListEqual(
+                list(attentions[0].shape[-3:]),
+                [
+                    self.model_tester.num_attention_heads,
+                    self.model_tester.num_feature_levels,
+                    self.model_tester.encoder_n_points,
+                ],
+            )
+            out_len = len(outputs)
+
+            correct_outlen = 10
+
+            # loss is at first position
+            if "labels" in inputs_dict:
+                correct_outlen += 1  # loss is added to beginning
+            # Object Detection model returns pred_logits and pred_boxes
+            if model_class.__name__ == "GroundingDinoForObjectDetection":
+                correct_outlen += 2
+
+            self.assertEqual(out_len, correct_outlen)
+
+            # decoder attentions
+            decoder_attentions = outputs.decoder_attentions[0]
+            self.assertIsInstance(decoder_attentions, (list, tuple))
+            self.assertEqual(len(decoder_attentions), self.model_tester.num_hidden_layers)
+            self.assertListEqual(
+                list(decoder_attentions[0].shape[-3:]),
+                [self.model_tester.num_attention_heads, self.model_tester.num_queries, self.model_tester.num_queries],
+            )
+
+            # cross attentions
+            cross_attentions = outputs.decoder_attentions[-1]
+            self.assertIsInstance(cross_attentions, (list, tuple))
+            self.assertEqual(len(cross_attentions), self.model_tester.num_hidden_layers)
+            self.assertListEqual(
+                list(cross_attentions[0].shape[-3:]),
+                [
+                    self.model_tester.num_attention_heads,
+                    self.model_tester.num_feature_levels,
+                    self.model_tester.decoder_n_points,
+                ],
+            )
+
+            # Check attention is always last and order is fine
+            inputs_dict["output_attentions"] = True
+            inputs_dict["output_hidden_states"] = True
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            self.assertEqual(out_len + 3, len(outputs))
+
+            self_attentions = outputs.encoder_attentions[-1]
+
+            self.assertEqual(len(self_attentions), self.model_tester.num_hidden_layers)
+            self.assertListEqual(
+                list(self_attentions[0].shape[-3:]),
+                [
+                    self.model_tester.num_attention_heads,
+                    self.model_tester.num_feature_levels,
+                    self.model_tester.encoder_n_points,
+                ],
+            )
+
+    # overwrite since hidden_states are called encoder_text_hidden_states
+    def test_hidden_states_output(self):
+        def check_hidden_states_output(inputs_dict, config, model_class):
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            hidden_states = outputs.encoder_vision_hidden_states
+
+            expected_num_layers = getattr(
+                self.model_tester, "expected_num_hidden_layers", self.model_tester.num_hidden_layers + 1
+            )
+            self.assertEqual(len(hidden_states), expected_num_layers)
+
+            seq_len = self.model_tester.encoder_seq_length_vision
+
+            self.assertListEqual(
+                list(hidden_states[0].shape[-2:]),
+                [seq_len, self.model_tester.hidden_size],
+            )
+
+            hidden_states = outputs.encoder_text_hidden_states
+
+            self.assertEqual(len(hidden_states), expected_num_layers)
+
+            seq_len = self.model_tester.encoder_seq_length_text
+
+            self.assertListEqual(
+                list(hidden_states[0].shape[-2:]),
+                [seq_len, self.model_tester.hidden_size],
+            )
+
+            hidden_states = outputs.decoder_hidden_states
+
+            self.assertIsInstance(hidden_states, (list, tuple))
+            self.assertEqual(len(hidden_states), expected_num_layers)
+            seq_len = getattr(self.model_tester, "seq_length", None)
+            decoder_seq_length = getattr(self.model_tester, "decoder_seq_length", seq_len)
+
+            self.assertListEqual(
+                list(hidden_states[0].shape[-2:]),
+                [decoder_seq_length, self.model_tester.hidden_size],
+            )
+
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            inputs_dict["output_hidden_states"] = True
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+            # check that output_hidden_states also work using config
+            del inputs_dict["output_hidden_states"]
+            config.output_hidden_states = True
+
+            check_hidden_states_output(inputs_dict, config, model_class)
+
+    # removed retain_grad and grad on decoder_hidden_states, as queries don't require grad
+    def test_retain_grad_hidden_states_attentions(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        config.output_hidden_states = True
+        config.output_attentions = True
+
+        # no need to test all models as different heads yield the same functionality
+        model_class = self.all_model_classes[0]
+        model = model_class(config)
+        model.to(torch_device)
+
+        inputs = self._prepare_for_class(inputs_dict, model_class)
+
+        outputs = model(**inputs)
+
+        output = outputs[0]
+
+        encoder_hidden_states = outputs.encoder_vision_hidden_states[0]
+        encoder_attentions = outputs.encoder_attentions[0][0]
+        encoder_hidden_states.retain_grad()
+        encoder_attentions.retain_grad()
+
+        cross_attentions = outputs.decoder_attentions[-1][0]
+        cross_attentions.retain_grad()
+
+        output.flatten()[0].backward(retain_graph=True)
+
+        self.assertIsNotNone(encoder_hidden_states.grad)
+        self.assertIsNotNone(encoder_attentions.grad)
+        self.assertIsNotNone(cross_attentions.grad)
+
+    def test_forward_signature(self):
+        config, _ = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+            signature = inspect.signature(model.forward)
+            # signature.parameters is an OrderedDict => so arg_names order is deterministic
+            arg_names = [*signature.parameters.keys()]
+
+            expected_arg_names = ["pixel_values", "input_ids"]
+            self.assertListEqual(arg_names[: len(expected_arg_names)], expected_arg_names)
+
+    def test_different_timm_backbone(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        # let's pick a random timm backbone
+        config.backbone = "tf_mobilenetv3_small_075"
+        config.use_timm_backbone = True
+        config.backbone_config = None
+        config.backbone_kwargs = {"in_chans": 3, "out_indices": (2, 3, 4)}
+
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+            with torch.no_grad():
+                outputs = model(**self._prepare_for_class(inputs_dict, model_class))
+
+            if model_class.__name__ == "GroundingDinoForObjectDetection":
+                expected_shape = (
+                    self.model_tester.batch_size,
+                    self.model_tester.num_queries,
+                    config.max_text_len,
+                )
+                self.assertEqual(outputs.logits.shape, expected_shape)
+
+            self.assertTrue(outputs)
+
+    def test_initialization(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        configs_no_init = _config_zero_init(config)
+        for model_class in self.all_model_classes:
+            model = model_class(config=configs_no_init)
+            for name, param in model.named_parameters():
+                if param.requires_grad:
+                    if (
+                        "level_embed" in name
+                        or "sampling_offsets.bias" in name
+                        or "text_param" in name
+                        or "vision_param" in name
+                        or "value_proj" in name
+                        or "output_proj" in name
+                        or "reference_points" in name
+                    ):
+                        continue
+                    self.assertIn(
+                        ((param.data.mean() * 1e9).round() / 1e9).item(),
+                        [0.0, 1.0],
+                        msg=f"Parameter {name} of model {model_class} seems not properly initialized",
+                    )
+
+    # Copied from tests.models.deformable_detr.test_modeling_deformable_detr.DeformableDetrModelTest.test_two_stage_training with DeformableDetr->GroundingDino
+    def test_two_stage_training(self):
+        model_class = GroundingDinoForObjectDetection
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+        config.return_dict = True
+        config.two_stage = True
+        config.auxiliary_loss = True
+        config.with_box_refine = True
+
+        model = model_class(config)
+        model.to(torch_device)
+        model.train()
+        inputs = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
+        loss = model(**inputs).loss
+        loss.backward()
+
+    def test_tied_weights_keys(self):
+        config, _ = self.model_tester.prepare_config_and_inputs_for_common()
+        config.tie_word_embeddings = True
+        for model_class in self.all_model_classes:
+            model_tied = model_class(config)
+
+            ptrs = collections.defaultdict(list)
+            for name, tensor in model_tied.state_dict().items():
+                ptrs[id_tensor_storage(tensor)].append(name)
+
+            # These are all the pointers of shared tensors.
+            tied_params = [names for _, names in ptrs.items() if len(names) > 1]
+
+            tied_weight_keys = model_tied._tied_weights_keys if model_tied._tied_weights_keys is not None else []
+            # Detect we get a hit for each key
+            for key in tied_weight_keys:
+                if not any(re.search(key, p) for group in tied_params for p in group):
+                    raise ValueError(f"{key} is not a tied weight key for {model_class}.")
+
+            # Removed tied weights found from tied params -> there should only be one left after
+            for key in tied_weight_keys:
+                for i in range(len(tied_params)):
+                    tied_params[i] = [p for p in tied_params[i] if re.search(key, p) is None]
+
+            # GroundingDino when sharing weights also uses the shared ones in GroundingDinoDecoder
+            # Therefore, differently from DeformableDetr, we expect the group lens to be 2
+            # one for self.bbox_embed in GroundingDinoForObejectDetection and another one
+            # in the decoder
+            tied_params = [group for group in tied_params if len(group) > 2]
+            self.assertListEqual(
+                tied_params,
+                [],
+                f"Missing `_tied_weights_keys` for {model_class}: add all of {tied_params} except one.",
+            )
+
+
+# We will verify our results on an image of cute cats
+def prepare_img():
+    image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
+    return image
+
+
+def prepare_text():
+    text = "a cat."
+    return text
+
+
+@require_timm
+@require_vision
+@slow
+class GroundingDinoModelIntegrationTests(unittest.TestCase):
+    @cached_property
+    def default_processor(self):
+        return AutoProcessor.from_pretrained("IDEA-Research/grounding-dino-tiny") if is_vision_available() else None
+
+    def test_inference_object_detection_head(self):
+        model = GroundingDinoForObjectDetection.from_pretrained("IDEA-Research/grounding-dino-tiny").to(torch_device)
+
+        processor = self.default_processor
+        image = prepare_img()
+        text = prepare_text()
+        encoding = processor(images=image, text=text, return_tensors="pt").to(torch_device)
+
+        with torch.no_grad():
+            outputs = model(**encoding)
+
+        expected_shape_logits = torch.Size((1, model.config.num_queries, model.config.d_model))
+        self.assertEqual(outputs.logits.shape, expected_shape_logits)
+
+        expected_boxes = torch.tensor(
+            [[0.7674, 0.4136, 0.4572], [0.2566, 0.5463, 0.4760], [0.2585, 0.5442, 0.4641]]
+        ).to(torch_device)
+        expected_logits = torch.tensor(
+            [[-4.8913, -0.1900, -0.2161], [-4.9653, -0.3719, -0.3950], [-5.9599, -3.3765, -3.3104]]
+        ).to(torch_device)
+
+        self.assertTrue(torch.allclose(outputs.logits[0, :3, :3], expected_logits, atol=1e-3))
+
+        expected_shape_boxes = torch.Size((1, model.config.num_queries, 4))
+        self.assertEqual(outputs.pred_boxes.shape, expected_shape_boxes)
+        self.assertTrue(torch.allclose(outputs.pred_boxes[0, :3, :3], expected_boxes, atol=1e-4))
+
+        # verify postprocessing
+        results = processor.image_processor.post_process_object_detection(
+            outputs, threshold=0.35, target_sizes=[image.size[::-1]]
+        )[0]
+        expected_scores = torch.tensor([0.4526, 0.4082]).to(torch_device)
+        expected_slice_boxes = torch.tensor([344.8143, 23.1796, 637.4004, 373.8295]).to(torch_device)
+
+        self.assertEqual(len(results["scores"]), 2)
+        self.assertTrue(torch.allclose(results["scores"], expected_scores, atol=1e-3))
+        self.assertTrue(torch.allclose(results["boxes"][0, :], expected_slice_boxes, atol=1e-2))
+
+        # verify grounded postprocessing
+        expected_labels = ["a cat", "a cat"]
+        results = processor.post_process_grounded_object_detection(
+            outputs=outputs,
+            input_ids=encoding.input_ids,
+            box_threshold=0.35,
+            text_threshold=0.3,
+            target_sizes=[image.size[::-1]],
+        )[0]
+
+        self.assertTrue(torch.allclose(results["scores"], expected_scores, atol=1e-3))
+        self.assertTrue(torch.allclose(results["boxes"][0, :], expected_slice_boxes, atol=1e-2))
+        self.assertListEqual(results["labels"], expected_labels)
+
+    @require_torch_gpu
+    def test_inference_object_detection_head_equivalence_cpu_gpu(self):
+        processor = self.default_processor
+        image = prepare_img()
+        text = prepare_text()
+        encoding = processor(images=image, text=text, return_tensors="pt")
+
+        # 1. run model on CPU
+        model = GroundingDinoForObjectDetection.from_pretrained("IDEA-Research/grounding-dino-tiny")
+
+        with torch.no_grad():
+            cpu_outputs = model(**encoding)
+
+        # 2. run model on GPU
+        model.to("cuda")
+        encoding = encoding.to("cuda")
+        with torch.no_grad():
+            gpu_outputs = model(**encoding)
+
+        # 3. assert equivalence
+        for key in cpu_outputs.keys():
+            self.assertTrue(torch.allclose(cpu_outputs[key], gpu_outputs[key].cpu(), atol=1e-3))
+
+        expected_logits = torch.tensor(
+            [[-4.8915, -0.1900, -0.2161], [-4.9658, -0.3716, -0.3948], [-5.9596, -3.3763, -3.3103]]
+        )
+        self.assertTrue(torch.allclose(cpu_outputs.logits[0, :3, :3], expected_logits, atol=1e-3))
+
+        # assert postprocessing
+        results_cpu = processor.image_processor.post_process_object_detection(
+            cpu_outputs, threshold=0.35, target_sizes=[image.size[::-1]]
+        )[0]
+
+        result_gpu = processor.image_processor.post_process_object_detection(
+            gpu_outputs, threshold=0.35, target_sizes=[image.size[::-1]]
+        )[0]
+
+        self.assertTrue(torch.allclose(results_cpu["scores"], result_gpu["scores"].cpu(), atol=1e-3))
+        self.assertTrue(torch.allclose(results_cpu["boxes"], result_gpu["boxes"].cpu(), atol=1e-3))
--- a/tests/models/grounding_dino/test_processor_grounding_dino.py
+++ b/tests/models/grounding_dino/test_processor_grounding_dino.py
@@ -0,0 +1,253 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+import shutil
+import tempfile
+import unittest
+
+import numpy as np
+import pytest
+
+from transformers import BertTokenizer, BertTokenizerFast, GroundingDinoProcessor
+from transformers.models.bert.tokenization_bert import VOCAB_FILES_NAMES
+from transformers.testing_utils import require_torch, require_vision
+from transformers.utils import IMAGE_PROCESSOR_NAME, is_torch_available, is_vision_available
+
+
+if is_torch_available():
+    import torch
+
+    from transformers.models.grounding_dino.modeling_grounding_dino import GroundingDinoObjectDetectionOutput
+
+if is_vision_available():
+    from PIL import Image
+
+    from transformers import GroundingDinoImageProcessor
+
+
+@require_torch
+@require_vision
+class GroundingDinoProcessorTest(unittest.TestCase):
+    def setUp(self):
+        self.tmpdirname = tempfile.mkdtemp()
+
+        vocab_tokens = ["[UNK]","[CLS]","[SEP]","[PAD]","[MASK]","want","##want","##ed","wa","un","runn","##ing",",","low","lowest"]  # fmt: skip
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
+            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
+
+        image_processor_map = {
+            "do_resize": True,
+            "size": None,
+            "do_normalize": True,
+            "image_mean": [0.5, 0.5, 0.5],
+            "image_std": [0.5, 0.5, 0.5],
+            "do_rescale": True,
+            "rescale_factor": 1 / 255,
+            "do_pad": True,
+        }
+        self.image_processor_file = os.path.join(self.tmpdirname, IMAGE_PROCESSOR_NAME)
+        with open(self.image_processor_file, "w", encoding="utf-8") as fp:
+            json.dump(image_processor_map, fp)
+
+        self.batch_size = 7
+        self.num_queries = 5
+        self.embed_dim = 5
+        self.seq_length = 5
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.get_tokenizer with CLIP->Bert
+    def get_tokenizer(self, **kwargs):
+        return BertTokenizer.from_pretrained(self.tmpdirname, **kwargs)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.get_rust_tokenizer with CLIP->Bert
+    def get_rust_tokenizer(self, **kwargs):
+        return BertTokenizerFast.from_pretrained(self.tmpdirname, **kwargs)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.get_image_processor with CLIP->GroundingDino
+    def get_image_processor(self, **kwargs):
+        return GroundingDinoImageProcessor.from_pretrained(self.tmpdirname, **kwargs)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.tearDown
+    def tearDown(self):
+        shutil.rmtree(self.tmpdirname)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.prepare_image_inputs
+    def prepare_image_inputs(self):
+        """This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
+        or a list of PyTorch tensors if one specifies torchify=True.
+        """
+
+        image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
+
+        image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
+
+        return image_inputs
+
+    def get_fake_grounding_dino_output(self):
+        torch.manual_seed(42)
+        return GroundingDinoObjectDetectionOutput(
+            pred_boxes=torch.rand(self.batch_size, self.num_queries, 4),
+            logits=torch.rand(self.batch_size, self.num_queries, self.embed_dim),
+        )
+
+    def get_fake_grounding_dino_input_ids(self):
+        input_ids = torch.tensor([101, 1037, 4937, 1012, 102])
+        return torch.stack([input_ids] * self.batch_size, dim=0)
+
+    def test_post_process_grounded_object_detection(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        grounding_dino_output = self.get_fake_grounding_dino_output()
+        grounding_dino_input_ids = self.get_fake_grounding_dino_input_ids()
+
+        post_processed = processor.post_process_grounded_object_detection(
+            grounding_dino_output, grounding_dino_input_ids
+        )
+
+        self.assertEqual(len(post_processed), self.batch_size)
+        self.assertEqual(list(post_processed[0].keys()), ["scores", "labels", "boxes"])
+        self.assertEqual(post_processed[0]["boxes"].shape, (self.num_queries, 4))
+        self.assertEqual(post_processed[0]["scores"].shape, (self.num_queries,))
+
+        expected_scores = torch.tensor([0.7050, 0.7222, 0.7222, 0.6829, 0.7220])
+        self.assertTrue(torch.allclose(post_processed[0]["scores"], expected_scores, atol=1e-4))
+
+        expected_box_slice = torch.tensor([0.6908, 0.4354, 1.0737, 1.3947])
+        self.assertTrue(torch.allclose(post_processed[0]["boxes"][0], expected_box_slice, atol=1e-4))
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_save_load_pretrained_default with CLIP->GroundingDino,GroundingDinoTokenizer->BertTokenizer
+    def test_save_load_pretrained_default(self):
+        tokenizer_slow = self.get_tokenizer()
+        tokenizer_fast = self.get_rust_tokenizer()
+        image_processor = self.get_image_processor()
+
+        processor_slow = GroundingDinoProcessor(tokenizer=tokenizer_slow, image_processor=image_processor)
+        processor_slow.save_pretrained(self.tmpdirname)
+        processor_slow = GroundingDinoProcessor.from_pretrained(self.tmpdirname, use_fast=False)
+
+        processor_fast = GroundingDinoProcessor(tokenizer=tokenizer_fast, image_processor=image_processor)
+        processor_fast.save_pretrained(self.tmpdirname)
+        processor_fast = GroundingDinoProcessor.from_pretrained(self.tmpdirname)
+
+        self.assertEqual(processor_slow.tokenizer.get_vocab(), tokenizer_slow.get_vocab())
+        self.assertEqual(processor_fast.tokenizer.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertEqual(tokenizer_slow.get_vocab(), tokenizer_fast.get_vocab())
+        self.assertIsInstance(processor_slow.tokenizer, BertTokenizer)
+        self.assertIsInstance(processor_fast.tokenizer, BertTokenizerFast)
+
+        self.assertEqual(processor_slow.image_processor.to_json_string(), image_processor.to_json_string())
+        self.assertEqual(processor_fast.image_processor.to_json_string(), image_processor.to_json_string())
+        self.assertIsInstance(processor_slow.image_processor, GroundingDinoImageProcessor)
+        self.assertIsInstance(processor_fast.image_processor, GroundingDinoImageProcessor)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_save_load_pretrained_additional_features with CLIP->GroundingDino,GroundingDinoTokenizer->BertTokenizer
+    def test_save_load_pretrained_additional_features(self):
+        processor = GroundingDinoProcessor(tokenizer=self.get_tokenizer(), image_processor=self.get_image_processor())
+        processor.save_pretrained(self.tmpdirname)
+
+        tokenizer_add_kwargs = self.get_tokenizer(bos_token="(BOS)", eos_token="(EOS)")
+        image_processor_add_kwargs = self.get_image_processor(do_normalize=False, padding_value=1.0)
+
+        processor = GroundingDinoProcessor.from_pretrained(
+            self.tmpdirname, bos_token="(BOS)", eos_token="(EOS)", do_normalize=False, padding_value=1.0
+        )
+
+        self.assertEqual(processor.tokenizer.get_vocab(), tokenizer_add_kwargs.get_vocab())
+        self.assertIsInstance(processor.tokenizer, BertTokenizerFast)
+
+        self.assertEqual(processor.image_processor.to_json_string(), image_processor_add_kwargs.to_json_string())
+        self.assertIsInstance(processor.image_processor, GroundingDinoImageProcessor)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_image_processor with CLIP->GroundingDino
+    def test_image_processor(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        image_input = self.prepare_image_inputs()
+
+        input_image_proc = image_processor(image_input, return_tensors="np")
+        input_processor = processor(images=image_input, return_tensors="np")
+
+        for key in input_image_proc.keys():
+            self.assertAlmostEqual(input_image_proc[key].sum(), input_processor[key].sum(), delta=1e-2)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_tokenizer with CLIP->GroundingDino
+    def test_tokenizer(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        input_str = "lower newer"
+
+        encoded_processor = processor(text=input_str)
+
+        encoded_tok = tokenizer(input_str)
+
+        for key in encoded_tok.keys():
+            self.assertListEqual(encoded_tok[key], encoded_processor[key])
+
+    def test_processor(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        input_str = "lower newer"
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(text=input_str, images=image_input)
+
+        self.assertListEqual(
+            list(inputs.keys()), ["input_ids", "token_type_ids", "attention_mask", "pixel_values", "pixel_mask"]
+        )
+
+        # test if it raises when no input is passed
+        with pytest.raises(ValueError):
+            processor()
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_tokenizer_decode with CLIP->GroundingDino
+    def test_tokenizer_decode(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        predicted_ids = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]
+
+        decoded_processor = processor.batch_decode(predicted_ids)
+        decoded_tok = tokenizer.batch_decode(predicted_ids)
+
+        self.assertListEqual(decoded_tok, decoded_processor)
+
+    # Copied from tests.models.clip.test_processor_clip.CLIPProcessorTest.test_model_input_names with CLIP->GroundingDino
+    def test_model_input_names(self):
+        image_processor = self.get_image_processor()
+        tokenizer = self.get_tokenizer()
+
+        processor = GroundingDinoProcessor(tokenizer=tokenizer, image_processor=image_processor)
+
+        input_str = "lower newer"
+        image_input = self.prepare_image_inputs()
+
+        inputs = processor(text=input_str, images=image_input)
+
+        self.assertListEqual(list(inputs.keys()), processor.model_input_names)