This commit is contained in:
2025-10-09 16:47:16 +08:00
parent c8feb4deb5
commit e27e3f16bb
5248 changed files with 1778505 additions and 0 deletions

View File

@@ -0,0 +1,190 @@
# 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 unittest
import numpy as np
from transformers.image_utils import PILImageResampling
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import (
is_torch_available,
is_torchvision_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 transformers import EfficientNetImageProcessor
if is_torchvision_available():
from transformers import EfficientNetImageProcessorFast
class EfficientNetImageProcessorTester:
def __init__(
self,
parent,
batch_size=13,
num_channels=3,
image_size=18,
min_resolution=30,
max_resolution=400,
do_resize=True,
size=None,
do_normalize=True,
image_mean=[0.5, 0.5, 0.5],
image_std=[0.5, 0.5, 0.5],
do_rescale=True,
rescale_offset=True,
rescale_factor=1 / 127.5,
resample=PILImageResampling.BILINEAR, # NEAREST is too different between PIL and torchvision
):
size = size if size is not None else {"height": 18, "width": 18}
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_resize = do_resize
self.size = size
self.do_normalize = do_normalize
self.image_mean = image_mean
self.image_std = image_std
self.resample = resample
def prepare_image_processor_dict(self):
return {
"image_mean": self.image_mean,
"image_std": self.image_std,
"do_normalize": self.do_normalize,
"do_resize": self.do_resize,
"size": self.size,
"resample": self.resample,
}
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 EfficientNetImageProcessorTest(ImageProcessingTestMixin, unittest.TestCase):
image_processing_class = EfficientNetImageProcessor if is_vision_available() else None
fast_image_processing_class = EfficientNetImageProcessorFast if is_torchvision_available() else None
def setUp(self):
super().setUp()
self.image_processor_tester = EfficientNetImageProcessorTester(self)
@property
def image_processor_dict(self):
return self.image_processor_tester.prepare_image_processor_dict()
def test_image_processor_properties(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
self.assertTrue(hasattr(image_processing, "image_mean"))
self.assertTrue(hasattr(image_processing, "image_std"))
self.assertTrue(hasattr(image_processing, "do_normalize"))
self.assertTrue(hasattr(image_processing, "do_resize"))
self.assertTrue(hasattr(image_processing, "size"))
def test_image_processor_from_dict_with_kwargs(self):
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class.from_dict(self.image_processor_dict)
self.assertEqual(image_processor.size, {"height": 18, "width": 18})
image_processor = image_processing_class.from_dict(self.image_processor_dict, size=42)
self.assertEqual(image_processor.size, {"height": 42, "width": 42})
def test_rescale(self):
# EfficientNet optionally rescales between -1 and 1 instead of the usual 0 and 1
image = np.arange(0, 256, 1, dtype=np.uint8).reshape(1, 8, 32)
for image_processing_class in self.image_processor_list:
image_processor = image_processing_class(**self.image_processor_dict)
if image_processing_class == EfficientNetImageProcessorFast:
image = torch.from_numpy(image)
# Scale between [-1, 1] with rescale_factor 1/127.5 and rescale_offset=True
rescaled_image = image_processor.rescale(image, scale=1 / 127.5, offset=True)
expected_image = (image * (1 / 127.5)) - 1
self.assertTrue(torch.allclose(rescaled_image, expected_image))
# Scale between [0, 1] with rescale_factor 1/255 and rescale_offset=True
rescaled_image = image_processor.rescale(image, scale=1 / 255, offset=False)
expected_image = image / 255.0
self.assertTrue(torch.allclose(rescaled_image, expected_image))
else:
rescaled_image = image_processor.rescale(image, scale=1 / 127.5, dtype=np.float64)
expected_image = (image * (1 / 127.5)).astype(np.float64) - 1
self.assertTrue(np.allclose(rescaled_image, expected_image))
rescaled_image = image_processor.rescale(image, scale=1 / 255, offset=False, dtype=np.float64)
expected_image = (image / 255.0).astype(np.float64)
self.assertTrue(np.allclose(rescaled_image, expected_image))
@require_vision
@require_torch
def test_rescale_normalize(self):
if self.image_processing_class is None or self.fast_image_processing_class is None:
self.skipTest(reason="Skipping slow/fast equivalence test as one of the image processors is not defined")
image = torch.arange(0, 256, 1, dtype=torch.uint8).reshape(1, 8, 32).repeat(3, 1, 1)
image_mean_0 = (0.0, 0.0, 0.0)
image_std_0 = (1.0, 1.0, 1.0)
image_mean_1 = (0.5, 0.5, 0.5)
image_std_1 = (0.5, 0.5, 0.5)
image_processor_fast = self.fast_image_processing_class(**self.image_processor_dict)
# Rescale between [-1, 1] with rescale_factor=1/127.5 and rescale_offset=True. Then normalize
rescaled_normalized = image_processor_fast.rescale_and_normalize(
image, True, 1 / 127.5, True, image_mean_0, image_std_0, True
)
expected_image = (image * (1 / 127.5)) - 1
expected_image = (expected_image - torch.tensor(image_mean_0).view(3, 1, 1)) / torch.tensor(image_std_0).view(
3, 1, 1
)
self.assertTrue(torch.allclose(rescaled_normalized, expected_image, rtol=1e-3))
# Rescale between [0, 1] with rescale_factor=1/255 and rescale_offset=False. Then normalize
rescaled_normalized = image_processor_fast.rescale_and_normalize(
image, True, 1 / 255, True, image_mean_1, image_std_1, False
)
expected_image = image * (1 / 255.0)
expected_image = (expected_image - torch.tensor(image_mean_1).view(3, 1, 1)) / torch.tensor(image_std_1).view(
3, 1, 1
)
self.assertTrue(torch.allclose(rescaled_normalized, expected_image, rtol=1e-3))

View File

@@ -0,0 +1,263 @@
# Copyright 2023 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 PyTorch EfficientNet model."""
import unittest
from functools import cached_property
from transformers import EfficientNetConfig
from transformers.testing_utils import is_pipeline_test, require_torch, require_vision, slow, torch_device
from transformers.utils import is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from transformers import EfficientNetForImageClassification, EfficientNetModel
if is_vision_available():
from PIL import Image
from transformers import AutoImageProcessor
class EfficientNetModelTester:
def __init__(
self,
parent,
batch_size=13,
image_size=32,
num_channels=3,
kernel_sizes=[3, 3, 5],
in_channels=[32, 16, 24],
out_channels=[16, 24, 20],
strides=[1, 1, 2],
num_block_repeats=[1, 1, 2],
expand_ratios=[1, 6, 6],
is_training=True,
use_labels=True,
intermediate_size=37,
hidden_act="gelu",
num_labels=10,
):
self.parent = parent
self.batch_size = batch_size
self.image_size = image_size
self.num_channels = num_channels
self.kernel_sizes = kernel_sizes
self.in_channels = in_channels
self.out_channels = out_channels
self.strides = strides
self.num_block_repeats = num_block_repeats
self.expand_ratios = expand_ratios
self.is_training = is_training
self.hidden_act = hidden_act
self.num_labels = num_labels
self.use_labels = use_labels
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.num_labels)
config = self.get_config()
return config, pixel_values, labels
def get_config(self):
return EfficientNetConfig(
image_size=self.image_size,
num_channels=self.num_channels,
kernel_sizes=self.kernel_sizes,
in_channels=self.in_channels,
out_channels=self.out_channels,
strides=self.strides,
num_block_repeats=self.num_block_repeats,
expand_ratios=self.expand_ratios,
hidden_act=self.hidden_act,
num_labels=self.num_labels,
)
def create_and_check_model(self, config, pixel_values, labels):
model = EfficientNetModel(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
# expected last hidden states: B, C, H // 4, W // 4
self.parent.assertEqual(
result.last_hidden_state.shape,
(self.batch_size, config.hidden_dim, self.image_size // 4, self.image_size // 4),
)
def create_and_check_for_image_classification(self, config, pixel_values, labels):
model = EfficientNetForImageClassification(config)
model.to(torch_device)
model.eval()
result = model(pixel_values, labels=labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
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_torch
class EfficientNetModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
"""
Here we also overwrite some of the tests of test_modeling_common.py, as EfficientNet does not use input_ids, inputs_embeds,
attention_mask and seq_length.
"""
all_model_classes = (EfficientNetModel, EfficientNetForImageClassification) if is_torch_available() else ()
pipeline_model_mapping = (
{"image-feature-extraction": EfficientNetModel, "image-classification": EfficientNetForImageClassification}
if is_torch_available()
else {}
)
fx_compatible = False
test_pruning = False
test_resize_embeddings = False
test_head_masking = False
has_attentions = False
test_torch_exportable = True
def setUp(self):
self.model_tester = EfficientNetModelTester(self)
self.config_tester = ConfigTester(
self,
config_class=EfficientNetConfig,
has_text_modality=False,
hidden_size=37,
common_properties=["num_channels", "image_size", "hidden_dim"],
)
def test_config(self):
self.config_tester.run_common_tests()
@unittest.skip(reason="EfficientNet does not use inputs_embeds")
def test_inputs_embeds(self):
pass
@unittest.skip(reason="EfficientNet does not support input and output embeddings")
def test_model_get_set_embeddings(self):
pass
@unittest.skip(reason="EfficientNet does not use feedforward chunking")
def test_feed_forward_chunking(self):
pass
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_hidden_states_output(self):
def check_hidden_states_output(inputs_dict, config, model_class):
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
hidden_states = outputs.encoder_hidden_states if config.is_encoder_decoder else outputs.hidden_states
num_blocks = sum(config.num_block_repeats) * 4
self.assertEqual(len(hidden_states), num_blocks)
# EfficientNet's feature maps are of shape (batch_size, num_channels, height, width)
self.assertListEqual(
list(hidden_states[0].shape[-2:]),
[self.model_tester.image_size // 2, self.model_tester.image_size // 2],
)
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
inputs_dict["output_hidden_states"] = True
check_hidden_states_output(inputs_dict, config, model_class)
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
config.output_hidden_states = True
check_hidden_states_output(inputs_dict, config, model_class)
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)
@slow
def test_model_from_pretrained(self):
model_name = "google/efficientnet-b7"
model = EfficientNetModel.from_pretrained(model_name)
self.assertIsNotNone(model)
@is_pipeline_test
@require_vision
@slow
def test_pipeline_image_feature_extraction(self):
super().test_pipeline_image_feature_extraction()
@is_pipeline_test
@require_vision
@slow
def test_pipeline_image_feature_extraction_fp16(self):
super().test_pipeline_image_feature_extraction_fp16()
@is_pipeline_test
@require_vision
@slow
def test_pipeline_image_classification(self):
super().test_pipeline_image_classification()
# 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_torch
@require_vision
class EfficientNetModelIntegrationTest(unittest.TestCase):
@cached_property
def default_image_processor(self):
return AutoImageProcessor.from_pretrained("google/efficientnet-b7") if is_vision_available() else None
@slow
def test_inference_image_classification_head(self):
model = EfficientNetForImageClassification.from_pretrained("google/efficientnet-b7").to(torch_device)
image_processor = self.default_image_processor
image = prepare_img()
inputs = image_processor(images=image, return_tensors="pt").to(torch_device)
# forward pass
with torch.no_grad():
outputs = model(**inputs)
# verify the logits
expected_shape = torch.Size((1, 1000))
self.assertEqual(outputs.logits.shape, expected_shape)
expected_slice = torch.tensor([-0.2962, 0.4487, 0.4499]).to(torch_device)
torch.testing.assert_close(outputs.logits[0, :3], expected_slice, rtol=1e-4, atol=1e-4)