transformers/tests/models/perceiver/test_image_processing_perceiver.py

# 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.image_utils import PILImageResampling
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_vision_available():
    from PIL import Image


if is_torch_available():
    import torch


class PerceiverImageProcessingTester:
    def __init__(
        self,
        parent,
        batch_size=7,
        num_channels=3,
        num_images=1,
        image_size=18,
        min_resolution=30,
        max_resolution=40,
        do_center_crop=True,
        crop_size=None,
        do_resize=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],
        resample=PILImageResampling.BICUBIC,
    ):
        self.crop_size = crop_size if crop_size is not None else {"height": 256, "width": 256}
        self.size = size if size is not None else {"height": 224, "width": 224}
        self.parent = parent
        self.batch_size = batch_size
        self.num_channels = num_channels
        self.num_images = num_images
        self.image_size = image_size
        self.min_resolution = min_resolution
        self.max_resolution = max_resolution
        self.do_center_crop = do_center_crop
        self.do_resize = do_resize
        self.resample = resample
        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_center_crop": self.do_center_crop,
            "crop_size": self.crop_size,
            "do_resize": self.do_resize,
            "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,
            "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 PerceiverImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
    def setUp(self):
        super().setUp()
        self.image_processor_tester = PerceiverImageProcessingTester(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_processing_classes.values():
            image_processing = image_processing_class(**self.image_processor_dict)
            self.assertTrue(hasattr(image_processing, "do_center_crop"))
            self.assertTrue(hasattr(image_processing, "crop_size"))
            self.assertTrue(hasattr(image_processing, "do_resize"))
            self.assertTrue(hasattr(image_processing, "size"))
            self.assertTrue(hasattr(image_processing, "resample"))
            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_call_numpy(self):
        for image_processing_class in self.image_processing_classes.values():
            # Initialize image_processing
            image_processing = 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 sample_images in image_inputs:
                for image in sample_images:
                    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_numpy_4_channels(self):
        # Idefics3 always processes images as RGB, so it always returns images with 3 channels
        for image_processing_class in self.image_processing_classes.values():
            # Initialize image_processing
            image_processor_dict = self.image_processor_dict
            image_processing = image_processing_class(**image_processor_dict)
            # create random numpy tensors
            image_inputs = self.image_processor_tester.prepare_image_inputs(equal_resolution=False, numpify=True)

            for sample_images in image_inputs:
                for image in sample_images:
                    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_pil(self):
        for image_processing_class in self.image_processing_classes.values():
            # Initialize image_processing
            image_processing = 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_pytorch(self):
        for image_processing_class in self.image_processing_classes.values():
            # Initialize image_processing
            image_processing = 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 images in image_inputs:
                for image in images:
                    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),
            )