Spaces:
Paused
Paused
File size: 10,886 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 |
# coding=utf-8
# Copyright 2022 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 TensorFlow DeiT model. """
from __future__ import annotations
import inspect
import unittest
import numpy as np
from transformers import DeiTConfig
from transformers.testing_utils import require_tf, require_vision, slow
from transformers.utils import cached_property, is_tf_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_tf_available():
import tensorflow as tf
from transformers import (
TFDeiTForImageClassification,
TFDeiTForImageClassificationWithTeacher,
TFDeiTForMaskedImageModeling,
TFDeiTModel,
)
from transformers.models.deit.modeling_tf_deit import TF_DEIT_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import DeiTImageProcessor
class TFDeiTModelTester:
def __init__(
self,
parent,
batch_size=13,
image_size=30,
patch_size=2,
num_channels=3,
is_training=True,
use_labels=True,
hidden_size=32,
num_hidden_layers=2,
num_attention_heads=4,
intermediate_size=37,
hidden_act="gelu",
hidden_dropout_prob=0.1,
attention_probs_dropout_prob=0.1,
type_sequence_label_size=10,
initializer_range=0.02,
num_labels=3,
scope=None,
encoder_stride=2,
):
self.parent = parent
self.batch_size = batch_size
self.image_size = image_size
self.patch_size = patch_size
self.num_channels = num_channels
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.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.scope = scope
self.encoder_stride = encoder_stride
# in DeiT, the seq length equals the number of patches + 2 (we add 2 for the [CLS] and distilation tokens)
num_patches = (image_size // patch_size) ** 2
self.seq_length = num_patches + 2
def prepare_config_and_inputs(self):
pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
labels = None
if self.use_labels:
labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
config = self.get_config()
return config, pixel_values, labels
def get_config(self):
return DeiTConfig(
image_size=self.image_size,
patch_size=self.patch_size,
num_channels=self.num_channels,
hidden_size=self.hidden_size,
num_hidden_layers=self.num_hidden_layers,
num_attention_heads=self.num_attention_heads,
intermediate_size=self.intermediate_size,
hidden_act=self.hidden_act,
hidden_dropout_prob=self.hidden_dropout_prob,
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
is_decoder=False,
initializer_range=self.initializer_range,
encoder_stride=self.encoder_stride,
)
def create_and_check_model(self, config, pixel_values, labels):
model = TFDeiTModel(config=config)
result = model(pixel_values)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_for_masked_image_modeling(self, config, pixel_values, labels):
model = TFDeiTForMaskedImageModeling(config=config)
result = model(pixel_values)
self.parent.assertEqual(
result.reconstruction.shape, (self.batch_size, self.num_channels, self.image_size, self.image_size)
)
# test greyscale images
config.num_channels = 1
model = TFDeiTForMaskedImageModeling(config)
pixel_values = floats_tensor([self.batch_size, 1, self.image_size, self.image_size])
result = model(pixel_values)
self.parent.assertEqual(result.reconstruction.shape, (self.batch_size, 1, self.image_size, self.image_size))
def create_and_check_for_image_classification(self, config, pixel_values, labels):
config.num_labels = self.type_sequence_label_size
model = TFDeiTForImageClassification(config)
result = model(pixel_values, labels=labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size))
# test greyscale images
config.num_channels = 1
model = TFDeiTForImageClassification(config)
pixel_values = floats_tensor([self.batch_size, 1, self.image_size, self.image_size])
result = model(pixel_values, labels=labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
config, pixel_values, labels = config_and_inputs
inputs_dict = {"pixel_values": pixel_values}
return config, inputs_dict
@require_tf
class TFDeiTModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
"""
Here we also overwrite some of the tests of test_modeling_tf_common.py, as DeiT does not use input_ids, inputs_embeds,
attention_mask and seq_length.
"""
all_model_classes = (
(
TFDeiTModel,
TFDeiTForImageClassification,
TFDeiTForImageClassificationWithTeacher,
TFDeiTForMaskedImageModeling,
)
if is_tf_available()
else ()
)
pipeline_model_mapping = (
{
"feature-extraction": TFDeiTModel,
"image-classification": (TFDeiTForImageClassification, TFDeiTForImageClassificationWithTeacher),
}
if is_tf_available()
else {}
)
test_pruning = False
test_resize_embeddings = False
test_head_masking = False
test_onnx = False
def setUp(self):
self.model_tester = TFDeiTModelTester(self)
self.config_tester = ConfigTester(self, config_class=DeiTConfig, has_text_modality=False, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
@unittest.skip(reason="DeiT does not use inputs_embeds")
def test_inputs_embeds(self):
pass
def test_model_common_attributes(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
self.assertIsInstance(model.get_input_embeddings(), (tf.keras.layers.Layer))
x = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(x, tf.keras.layers.Dense))
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.call)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = ["pixel_values"]
self.assertListEqual(arg_names[:1], expected_arg_names)
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_for_masked_image_modeling(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_masked_image_modeling(*config_and_inputs)
def test_for_image_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*config_and_inputs)
# special case for DeiTForImageClassificationWithTeacher model
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 "labels" in inputs_dict and "labels" not in inspect.signature(model_class.call).parameters:
del inputs_dict["labels"]
return inputs_dict
@slow
def test_model_from_pretrained(self):
for model_name in TF_DEIT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = TFDeiTModel.from_pretrained(model_name)
self.assertIsNotNone(model)
# 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
@require_tf
@require_vision
class DeiTModelIntegrationTest(unittest.TestCase):
@cached_property
def default_image_processor(self):
return (
DeiTImageProcessor.from_pretrained("facebook/deit-base-distilled-patch16-224")
if is_vision_available()
else None
)
@slow
def test_inference_image_classification_head(self):
model = TFDeiTForImageClassificationWithTeacher.from_pretrained("facebook/deit-base-distilled-patch16-224")
image_processor = self.default_image_processor
image = prepare_img()
inputs = image_processor(images=image, return_tensors="tf")
# forward pass
outputs = model(**inputs)
# verify the logits
expected_shape = tf.TensorShape((1, 1000))
self.assertEqual(outputs.logits.shape, expected_shape)
expected_slice = tf.constant([-1.0266, 0.1912, -1.2861])
self.assertTrue(np.allclose(outputs.logits[0, :3], expected_slice, atol=1e-4))
|