transformers/tests/test_processor_vision_text_dual_encoder.py

# Copyright 2021 The HuggingFace 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.

import json
import os
import shutil
import tempfile
import unittest

import numpy as np

from transformers import BertTokenizerFast
from transformers.file_utils import FEATURE_EXTRACTOR_NAME, is_vision_available
from transformers.models.bert.tokenization_bert import VOCAB_FILES_NAMES, BertTokenizer
from transformers.testing_utils import require_tokenizers, require_vision


if is_vision_available():
    from PIL import Image

    from transformers import VisionTextDualEncoderProcessor, ViTFeatureExtractor


@require_tokenizers
@require_vision
class VisionTextDualEncoderProcessorTest(unittest.TestCase):
    def setUp(self):
        self.tmpdirname = tempfile.mkdtemp()

        # fmt: off
        vocab_tokens = ["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]", "want", "##want", "##ed", "wa", "un", "runn", "##ing", ",", "low", "lowest"]
        # fmt: on
        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))

        feature_extractor_map = {
            "do_resize": True,
            "size": 18,
            "do_normalize": True,
            "image_mean": [0.5, 0.5, 0.5],
            "image_std": [0.5, 0.5, 0.5],
        }
        self.feature_extractor_file = os.path.join(self.tmpdirname, FEATURE_EXTRACTOR_NAME)
        with open(self.feature_extractor_file, "w", encoding="utf-8") as fp:
            json.dump(feature_extractor_map, fp)

    def get_tokenizer(self, **kwargs):
        return BertTokenizer.from_pretrained(self.tmpdirname, **kwargs)

    def get_feature_extractor(self, **kwargs):
        return ViTFeatureExtractor.from_pretrained(self.tmpdirname, **kwargs)

    def tearDown(self):
        shutil.rmtree(self.tmpdirname)

    def prepare_image_inputs(self):
        """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.
        """

        image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]

        image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]

        return image_inputs

    def test_save_load_pretrained_default(self):
        tokenizer = self.get_tokenizer()
        feature_extractor = self.get_feature_extractor()

        processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)

        processor.save_pretrained(self.tmpdirname)
        processor = VisionTextDualEncoderProcessor.from_pretrained(self.tmpdirname)

        self.assertEqual(processor.tokenizer.get_vocab(), tokenizer.get_vocab())
        self.assertIsInstance(processor.tokenizer, (BertTokenizer, BertTokenizerFast))

        self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor.to_json_string())
        self.assertIsInstance(processor.feature_extractor, ViTFeatureExtractor)

    def test_save_load_pretrained_additional_features(self):
        processor = VisionTextDualEncoderProcessor(
            tokenizer=self.get_tokenizer(), feature_extractor=self.get_feature_extractor()
        )
        processor.save_pretrained(self.tmpdirname)

        tokenizer_add_kwargs = self.get_tokenizer(bos_token="(BOS)", eos_token="(EOS)")
        feature_extractor_add_kwargs = self.get_feature_extractor(do_normalize=False, padding_value=1.0)

        processor = VisionTextDualEncoderProcessor.from_pretrained(
            self.tmpdirname, bos_token="(BOS)", eos_token="(EOS)", do_normalize=False, padding_value=1.0
        )

        self.assertEqual(processor.tokenizer.get_vocab(), tokenizer_add_kwargs.get_vocab())
        self.assertIsInstance(processor.tokenizer, (BertTokenizer, BertTokenizerFast))

        self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
        self.assertIsInstance(processor.feature_extractor, ViTFeatureExtractor)

    def test_feature_extractor(self):
        feature_extractor = self.get_feature_extractor()
        tokenizer = self.get_tokenizer()

        processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)

        image_input = self.prepare_image_inputs()

        input_feat_extract = feature_extractor(image_input, return_tensors="np")
        input_processor = processor(images=image_input, return_tensors="np")

        for key in input_feat_extract.keys():
            self.assertAlmostEqual(input_feat_extract[key].sum(), input_processor[key].sum(), delta=1e-2)

    def test_tokenizer(self):
        feature_extractor = self.get_feature_extractor()
        tokenizer = self.get_tokenizer()

        processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)

        input_str = "lower newer"

        encoded_processor = processor(text=input_str)

        encoded_tok = tokenizer(input_str)

        for key in encoded_tok.keys():
            self.assertListEqual(encoded_tok[key], encoded_processor[key])

    def test_processor(self):
        feature_extractor = self.get_feature_extractor()
        tokenizer = self.get_tokenizer()

        processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)

        input_str = "lower newer"
        image_input = self.prepare_image_inputs()

        inputs = processor(text=input_str, images=image_input)

        self.assertListEqual(list(inputs.keys()), ["input_ids", "token_type_ids", "attention_mask", "pixel_values"])

        # test if it raises when no input is passed
        with self.assertRaises(ValueError):
            processor()

    def test_tokenizer_decode(self):
        feature_extractor = self.get_feature_extractor()
        tokenizer = self.get_tokenizer()

        processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)

        predicted_ids = [[1, 4, 5, 8, 1, 0, 8], [3, 4, 3, 1, 1, 8, 9]]

        decoded_processor = processor.batch_decode(predicted_ids)
        decoded_tok = tokenizer.batch_decode(predicted_ids)

        self.assertListEqual(decoded_tok, decoded_processor)