File size: 10,680 Bytes
b381930 |
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 |
import shutil
import tempfile
import unittest
import torch
from transformers import GemmaTokenizer
from transformers.models.colpali.processing_colpali import ColPaliProcessor
from transformers.testing_utils import get_tests_dir, require_torch, require_vision
from transformers.utils import is_vision_available
from ...test_processing_common import ProcessorTesterMixin
if is_vision_available():
from transformers import (
ColPaliProcessor,
PaliGemmaProcessor,
SiglipImageProcessor,
)
SAMPLE_VOCAB = get_tests_dir("fixtures/test_sentencepiece.model")
@require_vision
class ColPaliProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor_class = ColPaliProcessor
@classmethod
def setUpClass(cls):
cls.tmpdirname = tempfile.mkdtemp()
image_processor = SiglipImageProcessor.from_pretrained("google/siglip-so400m-patch14-384")
image_processor.image_seq_length = 0
tokenizer = GemmaTokenizer(SAMPLE_VOCAB, keep_accents=True)
processor = PaliGemmaProcessor(image_processor=image_processor, tokenizer=tokenizer)
processor.save_pretrained(cls.tmpdirname)
@classmethod
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
@require_torch
@require_vision
def test_process_images(self):
# Processor configuration
image_input = self.prepare_image_inputs()
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer", max_length=112, padding="max_length")
image_processor.image_seq_length = 14
# Get the processor
processor = self.processor_class(
tokenizer=tokenizer,
image_processor=image_processor,
)
# Process the image
batch_feature = processor.process_images(images=image_input, return_tensors="pt")
# Assertions
self.assertIn("pixel_values", batch_feature)
self.assertEqual(batch_feature["pixel_values"].shape, torch.Size([1, 3, 384, 384]))
@require_torch
@require_vision
def test_process_queries(self):
# Inputs
queries = [
"Is attention really all you need?",
"Are Benjamin, Antoine, Merve, and Jo best friends?",
]
# Processor configuration
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer", max_length=112, padding="max_length")
image_processor.image_seq_length = 14
# Get the processor
processor = self.processor_class(
tokenizer=tokenizer,
image_processor=image_processor,
)
# Process the image
batch_feature = processor.process_queries(text=queries, return_tensors="pt")
# Assertions
self.assertIn("input_ids", batch_feature)
self.assertIsInstance(batch_feature["input_ids"], torch.Tensor)
self.assertEqual(batch_feature["input_ids"].shape[0], len(queries))
# The following tests are overwritten as ColPaliProcessor can only take one of images or text as input at a time
def test_tokenizer_defaults_preserved_by_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
input_str = self.prepare_text_inputs()
inputs = processor(text=input_str, return_tensors="pt")
self.assertEqual(inputs[self.text_input_name].shape[-1], 117)
def test_image_processor_defaults_preserved_by_image_kwargs(self):
"""
We use do_rescale=True, rescale_factor=-1 to ensure that image_processor kwargs are preserved in the processor.
We then check that the mean of the pixel_values is less than or equal to 0 after processing.
Since the original pixel_values are in [0, 255], this is a good indicator that the rescale_factor is indeed applied.
"""
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor_components["image_processor"] = self.get_component(
"image_processor", do_rescale=True, rescale_factor=-1
)
processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
image_input = self.prepare_image_inputs()
inputs = processor(images=image_input, return_tensors="pt")
self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
def test_kwargs_overrides_default_tokenizer_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor_components["tokenizer"] = self.get_component("tokenizer", padding="longest")
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
input_str = self.prepare_text_inputs()
inputs = processor(text=input_str, return_tensors="pt", max_length=112, padding="max_length")
self.assertEqual(inputs[self.text_input_name].shape[-1], 112)
def test_kwargs_overrides_default_image_processor_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor_components["image_processor"] = self.get_component(
"image_processor", do_rescale=True, rescale_factor=1
)
processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
image_input = self.prepare_image_inputs()
inputs = processor(images=image_input, do_rescale=True, rescale_factor=-1, return_tensors="pt")
self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
def test_unstructured_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
input_str = self.prepare_text_inputs()
inputs = processor(
text=input_str,
return_tensors="pt",
do_rescale=True,
rescale_factor=-1,
padding="max_length",
max_length=76,
)
self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
def test_unstructured_kwargs_batched(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
image_input = self.prepare_image_inputs(batch_size=2)
inputs = processor(
images=image_input,
return_tensors="pt",
do_rescale=True,
rescale_factor=-1,
padding="longest",
max_length=76,
)
self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
def test_doubly_passed_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
image_input = self.prepare_image_inputs()
with self.assertRaises(ValueError):
_ = processor(
images=image_input,
images_kwargs={"do_rescale": True, "rescale_factor": -1},
do_rescale=True,
return_tensors="pt",
)
def test_structured_kwargs_nested(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
input_str = self.prepare_text_inputs()
# Define the kwargs for each modality
all_kwargs = {
"common_kwargs": {"return_tensors": "pt"},
"images_kwargs": {"do_rescale": True, "rescale_factor": -1},
"text_kwargs": {"padding": "max_length", "max_length": 76},
}
inputs = processor(text=input_str, **all_kwargs)
self.skip_processor_without_typed_kwargs(processor)
self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
def test_structured_kwargs_nested_from_dict(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
processor_components = self.prepare_components()
processor = self.processor_class(**processor_components)
self.skip_processor_without_typed_kwargs(processor)
image_input = self.prepare_image_inputs()
# Define the kwargs for each modality
all_kwargs = {
"common_kwargs": {"return_tensors": "pt"},
"images_kwargs": {"do_rescale": True, "rescale_factor": -1},
"text_kwargs": {"padding": "max_length", "max_length": 76},
}
inputs = processor(images=image_input, **all_kwargs)
self.assertEqual(inputs[self.text_input_name].shape[-1], 76)
|