SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Add DETR (#11653)

Browse Source 
* Squash all commits of modeling_detr_v7 branch into one

* Improve docs

* Fix tests

* Style

* Improve docs some more and fix most tests

* Fix slow tests of ViT, DeiT and DETR

* Improve replacement of batch norm

* Restructure timm backbone forward

* Make DetrForSegmentation support any timm backbone

* Fix name of output

* Address most comments by @LysandreJik

* Give better names for variables

* Conditional imports + timm in setup.py

* Address additional comments by @sgugger

* Make style, add require_timm and require_vision to testsé

* Remove train_backbone attribute of DetrConfig, add methods to freeze/unfreeze backbone

* Add png files to fixtures

* Fix type hint

* Add timm to workflows

* Add `BatchNorm2d` to the weight initialization

* Fix retain_grad test

* Replace model checkpoints by Facebook namespace

* Fix name of checkpoint in test

* Add user-friendly message when scipy is not available

* Address most comments by @patrickvonplaten

* Remove return_intermediate_layers attribute of DetrConfig and simplify Joiner

* Better initialization

* Scipy is necessary to get sklearn metrics

* Rename TimmBackbone to DetrTimmConvEncoder and rename DetrJoiner to DetrConvModel

* Make style

* Improve docs and add 2 community notebooks

Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
This commit is contained in:
NielsRogge
2021-06-09 17:51:13 +02:00
committed by GitHub
parent d14e0af274
commit d3eacbb829
42 changed files with 5177 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
tests/test_feature_extraction_vit.py
View File

@@ -21,7 +21,7 @@ import numpy as np
from transformers.file_utils import is_torch_available, is_vision_available
from transformers.testing_utils import require_torch, require_vision
from .test_feature_extraction_common import FeatureExtractionSavingTestMixin
from .test_feature_extraction_common import FeatureExtractionSavingTestMixin, prepare_image_inputs
if is_torch_available():
@@ -69,36 +69,6 @@ class ViTFeatureExtractionTester(unittest.TestCase):
"size": self.size,
}
def prepare_inputs(self, equal_resolution=False, numpify=False, torchify=False):
"""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.
"""
assert not (numpify and torchify), "You cannot specify both numpy and PyTorch tensors at the same time"
if equal_resolution:
image_inputs = []
for i in range(self.batch_size):
image_inputs.append(
np.random.randint(
255, size=(self.num_channels, self.max_resolution, self.max_resolution), dtype=np.uint8
)
)
else:
image_inputs = []
for i in range(self.batch_size):
width, height = np.random.choice(np.arange(self.min_resolution, self.max_resolution), 2)
image_inputs.append(np.random.randint(255, size=(self.num_channels, width, height), dtype=np.uint8))
if not numpify and not torchify:
# PIL expects the channel dimension as last dimension
image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
if torchify:
image_inputs = [torch.from_numpy(x) for x in image_inputs]
return image_inputs
@require_torch
@require_vision
@@ -128,7 +98,7 @@ class ViTFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCa
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random PIL images
image_inputs = self.feature_extract_tester.prepare_inputs(equal_resolution=False)
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
for image in image_inputs:
self.assertIsInstance(image, Image.Image)
@@ -160,7 +130,7 @@ class ViTFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCa
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random numpy tensors
image_inputs = self.feature_extract_tester.prepare_inputs(equal_resolution=False, numpify=True)
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, numpify=True)
for image in image_inputs:
self.assertIsInstance(image, np.ndarray)
@@ -192,7 +162,7 @@ class ViTFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCa
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random PyTorch tensors
image_inputs = self.feature_extract_tester.prepare_inputs(equal_resolution=False, torchify=True)
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, torchify=True)
for image in image_inputs:
self.assertIsInstance(image, torch.Tensor)