mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-03 21:00:08 +06:00

* First draft * Make fixup * Make forward pass worké * Improve code * More improvements * More improvements * Make predictions match * More improvements * Improve image processor * Fix model tests * Add classic decoder * Convert classic decoder * Verify image processor * Fix classic decoder logits * Clean up * Add post_process_pose_estimation * Improve post_process_pose_estimation * Use AutoBackbone * Add support for MoE models * Fix tests, improve num_experts% * Improve variable names * Make fixup * More improvements * Improve post_process_pose_estimation * Compute centers and scales * Improve postprocessing * More improvements * Fix ViTPoseBackbone tests * Add docstrings, fix image processor tests * Update index * Use is_cv2_available * Add model to toctree * Add cv2 to doc tests * Remove script * Improve conversion script * Add coco_to_pascal_voc * Add box_to_center_and_scale to image_transforms * Update tests * Add integration test * Fix merge * Address comments * Replace numpy by pytorch, improve docstrings * Remove get_input_embeddings * Address comments * Move coco_to_pascal_voc * Address comment * Fix style * Address comments * Fix test * Address comment * Remove udp * Remove comment * [WIP] need to check if the numpy function is same as cv * add scipy affine_transform * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * refactor convert * add output_shape * add atol 5e-2 * Use hf_hub_download in conversion script * make box_to_center more applicable * skipt test_get_set_embedding * fix to accept array and fix CI * add co-contributor * make it to tensor type output * add torch * change to torch tensor * add more test * minor change * CI test change * import torch should be above ImageProcessor * make style * try not use torch in def * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/models/vitpose_backbone/configuration_vitpose_backbone.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/models/vitpose_backbone/modeling_vitpose_backbone.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/models/vitpose/modeling_vitpose.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * fix * fix * add caution * make more detail about dataset_index * Update src/transformers/models/vitpose/modeling_vitpose.py Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: Sangbum Daniel Choi <34004152+SangbumChoi@users.noreply.github.com> * add docs * Update docs/source/en/model_doc/vitpose.md * Update src/transformers/models/vitpose/configuration_vitpose.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Update src/transformers/__init__.py Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * Revert "Update src/transformers/__init__.py" This reverts commit7ffa504450
. * change name * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update tests/models/vitpose/test_modeling_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update docs/source/en/model_doc/vitpose.md Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/vitpose/modeling_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/vitpose_backbone/modeling_vitpose_backbone.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * move vitpose only function to image_processor * raise valueerror when using timm backbone * use out_indices * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * remove camel-case of def flip_back * rename vitposeEstimatorOutput * Update src/transformers/models/vitpose_backbone/modeling_vitpose_backbone.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * fix confused camelcase of MLP * remove in-place logic * clear scale description * make consistent batch format * docs update * formatting docstring * add batch tests * test docs change * Update src/transformers/models/vitpose/image_processing_vitpose.py * Update src/transformers/models/vitpose/configuration_vitpose.py * chagne ViT to Vit * change to enable MoE * make fix-copies * Update docs/source/en/model_doc/vitpose.md Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com> * extract udp * add more described docs * simple fix * change to accept target_size * make style * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * Update src/transformers/models/vitpose/configuration_vitpose.py Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * change to `verify_backbone_config_arguments` * Update docs/source/en/model_doc/vitpose.md Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com> * remove unnecessary copy * make config immutable * enable gradient checkpointing * update inappropriate docstring * linting docs * split function for visibility * make style * check isinstances * change to acceptable use_pretrained_backbone * make style * remove copy in docs * Update src/transformers/models/vitpose_backbone/modeling_vitpose_backbone.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * Update docs/source/en/model_doc/vitpose.md Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * Update src/transformers/models/vitpose/modeling_vitpose.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * simple fix + make style * change input config of activation function to string * Update docs/source/en/model_doc/vitpose.md Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * tmp docs * delete index.md * make fix-copies * simple fix * change conversion to sam2/mllama style * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * Update src/transformers/models/vitpose/image_processing_vitpose.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * refactor convert * add supervision * Update src/transformers/models/vitpose_backbone/modeling_vitpose_backbone.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * remove reduntant def * seperate code block for visualization * add validation for num_moe * final commit * add labels * [run-slow] vitpose, vitpose_backbone * Update src/transformers/models/vitpose/convert_vitpose_to_hf.py Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com> * enable all conversion * final commit * [run-slow] vitpose, vitpose_backbone * ruff check --fix * [run-slow] vitpose, vitpose_backbone * rename split module * [run-slow] vitpose, vitpose_backbone * fix pos_embed * Simplify init * Revert "fix pos_embed" This reverts commit2c56a4806e
. * refactor single loop * allow flag to enable custom model * efficiency of MoE to not use unused experts * make style * Fix range -> arange to avoid warning * Revert MOE router, a new one does not work * Fix postprocessing a bit (labels) * Fix type hint * Fix docs snippets * Fix links to checkpoints * Fix checkpoints in tests * Fix test * Add image to docs --------- Co-authored-by: Niels Rogge <nielsrogge@nielss-mbp.home> Co-authored-by: Niels Rogge <nielsrogge@Nielss-MacBook-Pro.local> Co-authored-by: sangbumchoi <danielsejong55@gmail.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: Pavel Iakubovskii <qubvel@gmail.com>
230 lines
9.7 KiB
Python
230 lines
9.7 KiB
Python
# 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 unittest
|
|
|
|
import numpy as np
|
|
|
|
from transformers.testing_utils import require_torch, require_vision
|
|
from transformers.utils import is_torch_available, is_vision_available
|
|
|
|
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 VitPoseImageProcessor
|
|
|
|
|
|
class VitPoseImageProcessingTester(unittest.TestCase):
|
|
def __init__(
|
|
self,
|
|
parent,
|
|
batch_size=7,
|
|
num_channels=3,
|
|
image_size=18,
|
|
min_resolution=30,
|
|
max_resolution=400,
|
|
do_affine_transform=True,
|
|
size=None,
|
|
do_rescale=True,
|
|
rescale_factor=1 / 255,
|
|
do_normalize=True,
|
|
image_mean=[0.5, 0.5, 0.5],
|
|
image_std=[0.5, 0.5, 0.5],
|
|
):
|
|
size = size if size is not None else {"height": 20, "width": 20}
|
|
self.parent = parent
|
|
self.batch_size = batch_size
|
|
self.num_channels = num_channels
|
|
self.image_size = image_size
|
|
self.min_resolution = min_resolution
|
|
self.max_resolution = max_resolution
|
|
self.do_affine_transform = do_affine_transform
|
|
self.size = size
|
|
self.do_rescale = do_rescale
|
|
self.rescale_factor = rescale_factor
|
|
self.do_normalize = do_normalize
|
|
self.image_mean = image_mean
|
|
self.image_std = image_std
|
|
|
|
def prepare_image_processor_dict(self):
|
|
return {
|
|
"do_affine_transform": self.do_affine_transform,
|
|
"size": self.size,
|
|
"do_rescale": self.do_rescale,
|
|
"rescale_factor": self.rescale_factor,
|
|
"do_normalize": self.do_normalize,
|
|
"image_mean": self.image_mean,
|
|
"image_std": self.image_std,
|
|
}
|
|
|
|
def expected_output_image_shape(self, images):
|
|
return self.num_channels, self.size["height"], self.size["width"]
|
|
|
|
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 VitPoseImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
|
|
image_processing_class = VitPoseImageProcessor if is_vision_available() else None
|
|
|
|
def setUp(self):
|
|
super().setUp()
|
|
self.image_processor_tester = VitPoseImageProcessingTester(self)
|
|
|
|
@property
|
|
def image_processor_dict(self):
|
|
return self.image_processor_tester.prepare_image_processor_dict()
|
|
|
|
def test_image_processor_properties(self):
|
|
image_processing = self.image_processing_class(**self.image_processor_dict)
|
|
self.assertTrue(hasattr(image_processing, "do_affine_transform"))
|
|
self.assertTrue(hasattr(image_processing, "size"))
|
|
self.assertTrue(hasattr(image_processing, "do_rescale"))
|
|
self.assertTrue(hasattr(image_processing, "rescale_factor"))
|
|
self.assertTrue(hasattr(image_processing, "do_normalize"))
|
|
self.assertTrue(hasattr(image_processing, "image_mean"))
|
|
self.assertTrue(hasattr(image_processing, "image_std"))
|
|
|
|
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, {"height": 20, "width": 20})
|
|
|
|
image_processor = self.image_processing_class.from_dict(
|
|
self.image_processor_dict, size={"height": 42, "width": 42}
|
|
)
|
|
self.assertEqual(image_processor.size, {"height": 42, "width": 42})
|
|
|
|
def test_call_pil(self):
|
|
# Initialize image_processing
|
|
image_processing = self.image_processing_class(**self.image_processor_dict)
|
|
# create random PIL images
|
|
image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False)
|
|
for image in image_inputs:
|
|
self.assertIsInstance(image, Image.Image)
|
|
|
|
# Test not batched input
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]]
|
|
encoded_images = image_processing(image_inputs[0], boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
|
|
self.assertEqual(tuple(encoded_images.shape), (2, *expected_output_image_shape))
|
|
|
|
# Test batched
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]] * self.image_processor_tester.batch_size
|
|
encoded_images = image_processing(image_inputs, boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
|
|
self.assertEqual(
|
|
tuple(encoded_images.shape), (self.image_processor_tester.batch_size * 2, *expected_output_image_shape)
|
|
)
|
|
|
|
def test_call_numpy(self):
|
|
# Initialize image_processing
|
|
image_processing = self.image_processing_class(**self.image_processor_dict)
|
|
# create random numpy tensors
|
|
image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)
|
|
for image in image_inputs:
|
|
self.assertIsInstance(image, np.ndarray)
|
|
|
|
# Test not batched input
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]]
|
|
encoded_images = image_processing(image_inputs[0], boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
|
|
self.assertEqual(tuple(encoded_images.shape), (2, *expected_output_image_shape))
|
|
|
|
# Test batched
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]] * self.image_processor_tester.batch_size
|
|
encoded_images = image_processing(image_inputs, boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
|
|
self.assertEqual(
|
|
tuple(encoded_images.shape), (self.image_processor_tester.batch_size * 2, *expected_output_image_shape)
|
|
)
|
|
|
|
def test_call_pytorch(self):
|
|
# Initialize image_processing
|
|
image_processing = 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)
|
|
|
|
for image in image_inputs:
|
|
self.assertIsInstance(image, torch.Tensor)
|
|
|
|
# Test not batched input
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]]
|
|
encoded_images = image_processing(image_inputs[0], boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
|
|
self.assertEqual(tuple(encoded_images.shape), (2, *expected_output_image_shape))
|
|
|
|
# Test batched
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]] * self.image_processor_tester.batch_size
|
|
encoded_images = image_processing(image_inputs, boxes=boxes, return_tensors="pt").pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
|
|
self.assertEqual(
|
|
tuple(encoded_images.shape), (self.image_processor_tester.batch_size * 2, *expected_output_image_shape)
|
|
)
|
|
|
|
def test_call_numpy_4_channels(self):
|
|
# Test that can process images which have an arbitrary number of channels
|
|
# Initialize image_processing
|
|
image_processor = self.image_processing_class(**self.image_processor_dict)
|
|
|
|
# create random numpy tensors
|
|
self.image_processor_tester.num_channels = 4
|
|
image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)
|
|
# Test not batched input
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]]
|
|
encoded_images = image_processor(
|
|
image_inputs[0],
|
|
boxes=boxes,
|
|
return_tensors="pt",
|
|
input_data_format="channels_last",
|
|
image_mean=0,
|
|
image_std=1,
|
|
).pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape([image_inputs[0]])
|
|
self.assertEqual(tuple(encoded_images.shape), (len(boxes[0]), *expected_output_image_shape))
|
|
|
|
# Test batched
|
|
boxes = [[[0, 0, 1, 1], [0.5, 0.5, 0.5, 0.5]]] * self.image_processor_tester.batch_size
|
|
encoded_images = image_processor(
|
|
image_inputs,
|
|
boxes=boxes,
|
|
return_tensors="pt",
|
|
input_data_format="channels_last",
|
|
image_mean=0,
|
|
image_std=1,
|
|
).pixel_values
|
|
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
|
|
self.assertEqual(
|
|
tuple(encoded_images.shape),
|
|
(self.image_processor_tester.batch_size * len(boxes[0]), *expected_output_image_shape),
|
|
)
|