Spaces:
Paused
Paused
File size: 12,216 Bytes
ee6e328 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 |
# coding=utf-8
# Copyright 2023 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 os
import tempfile
from transformers.testing_utils import check_json_file_has_correct_format, require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
if is_torch_available():
import numpy as np
import torch
if is_vision_available():
from PIL import Image
def prepare_image_inputs(
batch_size,
min_resolution,
max_resolution,
num_channels,
size_divisor=None,
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.
One can specify whether the images are of the same resolution or not.
"""
assert not (numpify and torchify), "You cannot specify both numpy and PyTorch tensors at the same time"
image_inputs = []
for i in range(batch_size):
if equal_resolution:
width = height = max_resolution
else:
# To avoid getting image width/height 0
if size_divisor is not None:
# If `size_divisor` is defined, the image needs to have width/size >= `size_divisor`
min_resolution = max(size_divisor, min_resolution)
width, height = np.random.choice(np.arange(min_resolution, max_resolution), 2)
image_inputs.append(np.random.randint(255, size=(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(image, 0, -1)) for image in image_inputs]
if torchify:
image_inputs = [torch.from_numpy(image) for image in image_inputs]
return image_inputs
def prepare_video(num_frames, num_channels, width=10, height=10, numpify=False, torchify=False):
"""This function prepares a video as a list of PIL images/NumPy arrays/PyTorch tensors."""
video = []
for i in range(num_frames):
video.append(np.random.randint(255, size=(num_channels, width, height), dtype=np.uint8))
if not numpify and not torchify:
# PIL expects the channel dimension as last dimension
video = [Image.fromarray(np.moveaxis(frame, 0, -1)) for frame in video]
if torchify:
video = [torch.from_numpy(frame) for frame in video]
return video
def prepare_video_inputs(
batch_size,
num_frames,
num_channels,
min_resolution,
max_resolution,
equal_resolution=False,
numpify=False,
torchify=False,
):
"""This function prepares a batch of videos: a list of list of PIL images, or a list of list of numpy arrays if
one specifies numpify=True, or a list of list of PyTorch tensors if one specifies torchify=True.
One can specify whether the videos are of the same resolution or not.
"""
assert not (numpify and torchify), "You cannot specify both numpy and PyTorch tensors at the same time"
video_inputs = []
for i in range(batch_size):
if equal_resolution:
width = height = max_resolution
else:
width, height = np.random.choice(np.arange(min_resolution, max_resolution), 2)
video = prepare_video(
num_frames=num_frames,
num_channels=num_channels,
width=width,
height=height,
numpify=numpify,
torchify=torchify,
)
video_inputs.append(video)
return video_inputs
class ImageProcessingTestMixin:
test_cast_dtype = None
def test_image_processor_to_json_string(self):
image_processor = self.image_processing_class(**self.image_processor_dict)
obj = json.loads(image_processor.to_json_string())
for key, value in self.image_processor_dict.items():
self.assertEqual(obj[key], value)
def test_image_processor_to_json_file(self):
image_processor_first = self.image_processing_class(**self.image_processor_dict)
with tempfile.TemporaryDirectory() as tmpdirname:
json_file_path = os.path.join(tmpdirname, "image_processor.json")
image_processor_first.to_json_file(json_file_path)
image_processor_second = self.image_processing_class.from_json_file(json_file_path)
self.assertEqual(image_processor_second.to_dict(), image_processor_first.to_dict())
def test_image_processor_from_and_save_pretrained(self):
image_processor_first = self.image_processing_class(**self.image_processor_dict)
with tempfile.TemporaryDirectory() as tmpdirname:
saved_file = image_processor_first.save_pretrained(tmpdirname)[0]
check_json_file_has_correct_format(saved_file)
image_processor_second = self.image_processing_class.from_pretrained(tmpdirname)
self.assertEqual(image_processor_second.to_dict(), image_processor_first.to_dict())
def test_init_without_params(self):
image_processor = self.image_processing_class()
self.assertIsNotNone(image_processor)
@require_torch
@require_vision
def test_cast_dtype_device(self):
if self.test_cast_dtype is not None:
# 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)
encoding = image_processor(image_inputs, return_tensors="pt")
# for layoutLM compatiblity
self.assertEqual(encoding.pixel_values.device, torch.device("cpu"))
self.assertEqual(encoding.pixel_values.dtype, torch.float32)
encoding = image_processor(image_inputs, return_tensors="pt").to(torch.float16)
self.assertEqual(encoding.pixel_values.device, torch.device("cpu"))
self.assertEqual(encoding.pixel_values.dtype, torch.float16)
encoding = image_processor(image_inputs, return_tensors="pt").to("cpu", torch.bfloat16)
self.assertEqual(encoding.pixel_values.device, torch.device("cpu"))
self.assertEqual(encoding.pixel_values.dtype, torch.bfloat16)
with self.assertRaises(TypeError):
_ = image_processor(image_inputs, return_tensors="pt").to(torch.bfloat16, "cpu")
# Try with text + image feature
encoding = image_processor(image_inputs, return_tensors="pt")
encoding.update({"input_ids": torch.LongTensor([[1, 2, 3], [4, 5, 6]])})
encoding = encoding.to(torch.float16)
self.assertEqual(encoding.pixel_values.device, torch.device("cpu"))
self.assertEqual(encoding.pixel_values.dtype, torch.float16)
self.assertEqual(encoding.input_ids.dtype, torch.long)
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
encoded_images = image_processing(image_inputs[0], 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), (1, *expected_output_image_shape))
# Test batched
encoded_images = image_processing(image_inputs, 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, *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
encoded_images = image_processing(image_inputs[0], 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), (1, *expected_output_image_shape))
# Test batched
encoded_images = image_processing(image_inputs, 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, *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
encoded_images = image_processing(image_inputs[0], 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), (1, *expected_output_image_shape))
# Test batched
expected_output_image_shape = self.image_processor_tester.expected_output_image_shape(image_inputs)
encoded_images = image_processing(image_inputs, return_tensors="pt").pixel_values
self.assertEqual(
tuple(encoded_images.shape),
(self.image_processor_tester.batch_size, *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
encoded_images = image_processor(
image_inputs[0],
return_tensors="pt",
input_data_format="channels_first",
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), (1, *expected_output_image_shape))
# Test batched
encoded_images = image_processor(
image_inputs,
return_tensors="pt",
input_data_format="channels_first",
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, *expected_output_image_shape)
)
|