252 lines
8.9 KiB
Python
252 lines
8.9 KiB
Python
# 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 PyTorch MobileNetV1 model."""
|
|
|
|
import unittest
|
|
|
|
from transformers import MobileNetV1Config
|
|
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
|
|
from transformers.utils import cached_property, 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 MobileNetV1ForImageClassification, MobileNetV1Model
|
|
|
|
|
|
if is_vision_available():
|
|
from PIL import Image
|
|
|
|
from transformers import MobileNetV1ImageProcessor
|
|
|
|
|
|
class MobileNetV1ConfigTester(ConfigTester):
|
|
def create_and_test_config_common_properties(self):
|
|
config = self.config_class(**self.inputs_dict)
|
|
self.parent.assertTrue(hasattr(config, "tf_padding"))
|
|
self.parent.assertTrue(hasattr(config, "depth_multiplier"))
|
|
|
|
|
|
class MobileNetV1ModelTester:
|
|
def __init__(
|
|
self,
|
|
parent,
|
|
batch_size=13,
|
|
num_channels=3,
|
|
image_size=32,
|
|
depth_multiplier=0.25,
|
|
min_depth=8,
|
|
tf_padding=True,
|
|
last_hidden_size=1024,
|
|
output_stride=32,
|
|
hidden_act="relu6",
|
|
classifier_dropout_prob=0.1,
|
|
initializer_range=0.02,
|
|
is_training=True,
|
|
use_labels=True,
|
|
num_labels=10,
|
|
scope=None,
|
|
):
|
|
self.parent = parent
|
|
self.batch_size = batch_size
|
|
self.num_channels = num_channels
|
|
self.image_size = image_size
|
|
self.depth_multiplier = depth_multiplier
|
|
self.min_depth = min_depth
|
|
self.tf_padding = tf_padding
|
|
self.last_hidden_size = int(last_hidden_size * depth_multiplier)
|
|
self.output_stride = output_stride
|
|
self.hidden_act = hidden_act
|
|
self.classifier_dropout_prob = classifier_dropout_prob
|
|
self.use_labels = use_labels
|
|
self.is_training = is_training
|
|
self.num_labels = num_labels
|
|
self.initializer_range = initializer_range
|
|
self.scope = scope
|
|
|
|
def prepare_config_and_inputs(self):
|
|
pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
|
|
|
|
labels = None
|
|
pixel_labels = None
|
|
if self.use_labels:
|
|
labels = ids_tensor([self.batch_size], self.num_labels)
|
|
pixel_labels = ids_tensor([self.batch_size, self.image_size, self.image_size], self.num_labels)
|
|
|
|
config = self.get_config()
|
|
|
|
return config, pixel_values, labels, pixel_labels
|
|
|
|
def get_config(self):
|
|
return MobileNetV1Config(
|
|
num_channels=self.num_channels,
|
|
image_size=self.image_size,
|
|
depth_multiplier=self.depth_multiplier,
|
|
min_depth=self.min_depth,
|
|
tf_padding=self.tf_padding,
|
|
hidden_act=self.hidden_act,
|
|
classifier_dropout_prob=self.classifier_dropout_prob,
|
|
initializer_range=self.initializer_range,
|
|
)
|
|
|
|
def create_and_check_model(self, config, pixel_values, labels, pixel_labels):
|
|
model = MobileNetV1Model(config=config)
|
|
model.to(torch_device)
|
|
model.eval()
|
|
result = model(pixel_values)
|
|
self.parent.assertEqual(
|
|
result.last_hidden_state.shape,
|
|
(
|
|
self.batch_size,
|
|
self.last_hidden_size,
|
|
self.image_size // self.output_stride,
|
|
self.image_size // self.output_stride,
|
|
),
|
|
)
|
|
|
|
def create_and_check_for_image_classification(self, config, pixel_values, labels, pixel_labels):
|
|
config.num_labels = self.num_labels
|
|
model = MobileNetV1ForImageClassification(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, pixel_labels = config_and_inputs
|
|
inputs_dict = {"pixel_values": pixel_values}
|
|
return config, inputs_dict
|
|
|
|
|
|
@require_torch
|
|
class MobileNetV1ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
|
|
"""
|
|
Here we also overwrite some of the tests of test_modeling_common.py, as MobileNetV1 does not use input_ids, inputs_embeds,
|
|
attention_mask and seq_length.
|
|
"""
|
|
|
|
all_model_classes = (MobileNetV1Model, MobileNetV1ForImageClassification) if is_torch_available() else ()
|
|
pipeline_model_mapping = (
|
|
{"image-feature-extraction": MobileNetV1Model, "image-classification": MobileNetV1ForImageClassification}
|
|
if is_torch_available()
|
|
else {}
|
|
)
|
|
|
|
test_pruning = False
|
|
test_resize_embeddings = False
|
|
test_head_masking = False
|
|
has_attentions = False
|
|
|
|
def setUp(self):
|
|
self.model_tester = MobileNetV1ModelTester(self)
|
|
self.config_tester = MobileNetV1ConfigTester(self, config_class=MobileNetV1Config, has_text_modality=False)
|
|
|
|
def test_config(self):
|
|
self.config_tester.run_common_tests()
|
|
|
|
@unittest.skip(reason="MobileNetV1 does not use inputs_embeds")
|
|
def test_inputs_embeds(self):
|
|
pass
|
|
|
|
@unittest.skip(reason="MobileNetV1 does not support input and output embeddings")
|
|
def test_model_common_attributes(self):
|
|
pass
|
|
|
|
@unittest.skip(reason="MobileNetV1 does not output attentions")
|
|
def test_attention_outputs(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.hidden_states
|
|
|
|
expected_num_stages = 26
|
|
self.assertEqual(len(hidden_states), expected_num_stages)
|
|
|
|
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/mobilenet_v1_1.0_224"
|
|
model = MobileNetV1Model.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_torch
|
|
@require_vision
|
|
class MobileNetV1ModelIntegrationTest(unittest.TestCase):
|
|
@cached_property
|
|
def default_image_processor(self):
|
|
return (
|
|
MobileNetV1ImageProcessor.from_pretrained("google/mobilenet_v1_1.0_224") if is_vision_available() else None
|
|
)
|
|
|
|
@slow
|
|
def test_inference_image_classification_head(self):
|
|
model = MobileNetV1ForImageClassification.from_pretrained("google/mobilenet_v1_1.0_224").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, 1001))
|
|
self.assertEqual(outputs.logits.shape, expected_shape)
|
|
|
|
expected_slice = torch.tensor([-4.1739, -1.1233, 3.1205]).to(torch_device)
|
|
|
|
self.assertTrue(torch.allclose(outputs.logits[0, :3], expected_slice, atol=1e-4))
|