🔴 🔴 🔴 Added segmentation maps support for DPT image processor (#34345)
* Added `segmentation_maps` support for DPT image processor * Added tests for dpt image processor * Moved preprocessing into separate functions * Added # Copied from statements * Fixed # Copied from statements * Added `segmentation_maps` support for DPT image processor * Added tests for dpt image processor * Moved preprocessing into separate functions * Added # Copied from statements * Fixed # Copied from statements
This commit is contained in:
Mikhail Moskovchenko
committed by
GitHub
GitHub
parent
a50befa9b9
commit
5450e7c84a
@@ -17,14 +17,20 @@
|
||||
import unittest
|
||||
|
||||
import numpy as np
|
||||
from datasets import load_dataset
|
||||
|
||||
from transformers.file_utils import is_vision_available
|
||||
from transformers.file_utils import is_torch_available, is_vision_available
|
||||
from transformers.testing_utils import require_torch, require_vision
|
||||
|
||||
from ...test_image_processing_common import ImageProcessingTestMixin, prepare_image_inputs
|
||||
|
||||
|
||||
if is_torch_available():
|
||||
import torch
|
||||
|
||||
if is_vision_available():
|
||||
from PIL import Image
|
||||
|
||||
from transformers import DPTImageProcessor
|
||||
|
||||
|
||||
@@ -42,6 +48,7 @@ class DPTImageProcessingTester(unittest.TestCase):
|
||||
do_normalize=True,
|
||||
image_mean=[0.5, 0.5, 0.5],
|
||||
image_std=[0.5, 0.5, 0.5],
|
||||
do_reduce_labels=False,
|
||||
):
|
||||
super().__init__()
|
||||
size = size if size is not None else {"height": 18, "width": 18}
|
||||
@@ -56,6 +63,7 @@ class DPTImageProcessingTester(unittest.TestCase):
|
||||
self.do_normalize = do_normalize
|
||||
self.image_mean = image_mean
|
||||
self.image_std = image_std
|
||||
self.do_reduce_labels = do_reduce_labels
|
||||
|
||||
def prepare_image_processor_dict(self):
|
||||
return {
|
||||
@@ -64,6 +72,7 @@ class DPTImageProcessingTester(unittest.TestCase):
|
||||
"do_normalize": self.do_normalize,
|
||||
"do_resize": self.do_resize,
|
||||
"size": self.size,
|
||||
"do_reduce_labels": self.do_reduce_labels,
|
||||
}
|
||||
|
||||
def expected_output_image_shape(self, images):
|
||||
@@ -81,6 +90,28 @@ class DPTImageProcessingTester(unittest.TestCase):
|
||||
)
|
||||
|
||||
|
||||
# Copied from transformers.tests.models.beit.test_image_processing_beit.prepare_semantic_single_inputs
|
||||
def prepare_semantic_single_inputs():
|
||||
dataset = load_dataset("hf-internal-testing/fixtures_ade20k", split="test", trust_remote_code=True)
|
||||
|
||||
image = Image.open(dataset[0]["file"])
|
||||
map = Image.open(dataset[1]["file"])
|
||||
|
||||
return image, map
|
||||
|
||||
|
||||
# Copied from transformers.tests.models.beit.test_image_processing_beit.prepare_semantic_batch_inputs
|
||||
def prepare_semantic_batch_inputs():
|
||||
ds = load_dataset("hf-internal-testing/fixtures_ade20k", split="test", trust_remote_code=True)
|
||||
|
||||
image1 = Image.open(ds[0]["file"])
|
||||
map1 = Image.open(ds[1]["file"])
|
||||
image2 = Image.open(ds[2]["file"])
|
||||
map2 = Image.open(ds[3]["file"])
|
||||
|
||||
return [image1, image2], [map1, map2]
|
||||
|
||||
|
||||
@require_torch
|
||||
@require_vision
|
||||
class DPTImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
||||
@@ -105,6 +136,7 @@ class DPTImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
||||
self.assertTrue(hasattr(image_processing, "rescale_factor"))
|
||||
self.assertTrue(hasattr(image_processing, "do_pad"))
|
||||
self.assertTrue(hasattr(image_processing, "size_divisor"))
|
||||
self.assertTrue(hasattr(image_processing, "do_reduce_labels"))
|
||||
|
||||
def test_image_processor_from_dict_with_kwargs(self):
|
||||
image_processor = self.image_processing_class.from_dict(self.image_processor_dict)
|
||||
@@ -138,3 +170,126 @@ class DPTImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
||||
pixel_values = image_processor(image, return_tensors="pt").pixel_values
|
||||
|
||||
self.assertEqual(list(pixel_values.shape), [1, 3, 512, 672])
|
||||
|
||||
# Copied from transformers.tests.models.beit.test_image_processing_beit.BeitImageProcessingTest.test_call_segmentation_maps
|
||||
def test_call_segmentation_maps(self):
|
||||
# Initialize image_processor
|
||||
image_processor = self.image_processing_class(**self.image_processor_dict)
|
||||
# create random PyTorch tensors
|
||||
image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, torchify=True)
|
||||
maps = []
|
||||
for image in image_inputs:
|
||||
self.assertIsInstance(image, torch.Tensor)
|
||||
maps.append(torch.zeros(image.shape[-2:]).long())
|
||||
|
||||
# Test not batched input
|
||||
encoding = image_processor(image_inputs[0], maps[0], return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
# Test batched
|
||||
encoding = image_processor(image_inputs, maps, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
self.image_processor_tester.batch_size,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
self.image_processor_tester.batch_size,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
# Test not batched input (PIL images)
|
||||
image, segmentation_map = prepare_semantic_single_inputs()
|
||||
|
||||
encoding = image_processor(image, segmentation_map, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
1,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
# Test batched input (PIL images)
|
||||
images, segmentation_maps = prepare_semantic_batch_inputs()
|
||||
|
||||
encoding = image_processor(images, segmentation_maps, return_tensors="pt")
|
||||
self.assertEqual(
|
||||
encoding["pixel_values"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.num_channels,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(
|
||||
encoding["labels"].shape,
|
||||
(
|
||||
2,
|
||||
self.image_processor_tester.size["height"],
|
||||
self.image_processor_tester.size["width"],
|
||||
),
|
||||
)
|
||||
self.assertEqual(encoding["labels"].dtype, torch.long)
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
# Copied from transformers.tests.models.beit.test_image_processing_beit.BeitImageProcessingTest.test_reduce_labels
|
||||
def test_reduce_labels(self):
|
||||
# Initialize image_processor
|
||||
image_processor = self.image_processing_class(**self.image_processor_dict)
|
||||
|
||||
# ADE20k has 150 classes, and the background is included, so labels should be between 0 and 150
|
||||
image, map = prepare_semantic_single_inputs()
|
||||
encoding = image_processor(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 150)
|
||||
|
||||
image_processor.do_reduce_labels = True
|
||||
encoding = image_processor(image, map, return_tensors="pt")
|
||||
self.assertTrue(encoding["labels"].min().item() >= 0)
|
||||
self.assertTrue(encoding["labels"].max().item() <= 255)
|
||||
|
||||
Reference in New Issue
Block a user