Spaces:
Running
Running
File size: 8,059 Bytes
9fd1204 |
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 |
import os
import tempfile
import unittest
from finetrainers.data import (
InMemoryDistributedDataPreprocessor,
PrecomputedDistributedDataPreprocessor,
VideoCaptionFilePairDataset,
initialize_preprocessor,
wrap_iterable_dataset_for_preprocessing,
)
from finetrainers.data.precomputation import PRECOMPUTED_DATA_DIR
from finetrainers.utils import find_files
from .utils import create_dummy_directory_structure
class PreprocessorFastTests(unittest.TestCase):
def setUp(self):
self.rank = 0
self.world_size = 1
self.num_items = 3
self.processor_fn = {
"latent": self._latent_processor_fn,
"condition": self._condition_processor_fn,
}
self.save_dir = tempfile.TemporaryDirectory()
directory_structure = [
"0.mp4",
"1.mp4",
"2.mp4",
"0.txt",
"1.txt",
"2.txt",
]
create_dummy_directory_structure(
directory_structure, self.save_dir, self.num_items, "a cat ruling the world", "mp4"
)
dataset = VideoCaptionFilePairDataset(self.save_dir.name, infinite=True)
dataset = wrap_iterable_dataset_for_preprocessing(
dataset,
dataset_type="video",
config={
"video_resolution_buckets": [[2, 32, 32]],
"reshape_mode": "bicubic",
},
)
self.dataset = dataset
def tearDown(self):
self.save_dir.cleanup()
@staticmethod
def _latent_processor_fn(**data):
video = data["video"]
video = video[:, :, :16, :16]
data["video"] = video
return data
@staticmethod
def _condition_processor_fn(**data):
caption = data["caption"]
caption = caption + " surrounded by mystical aura"
data["caption"] = caption
return data
def test_initialize_preprocessor(self):
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=False,
)
self.assertIsInstance(preprocessor, InMemoryDistributedDataPreprocessor)
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=True,
)
self.assertIsInstance(preprocessor, PrecomputedDistributedDataPreprocessor)
def test_in_memory_preprocessor_consume(self):
data_iterator = iter(self.dataset)
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=False,
)
condition_iterator = preprocessor.consume(
"condition", components={}, data_iterator=data_iterator, cache_samples=True
)
latent_iterator = preprocessor.consume(
"latent", components={}, data_iterator=data_iterator, use_cached_samples=True, drop_samples=True
)
self.assertFalse(preprocessor.requires_data)
for _ in range(self.num_items):
condition_item = next(condition_iterator)
latent_item = next(latent_iterator)
self.assertIn("caption", condition_item)
self.assertIn("video", latent_item)
self.assertEqual(condition_item["caption"], "a cat ruling the world surrounded by mystical aura")
self.assertEqual(latent_item["video"].shape[-2:], (16, 16))
self.assertTrue(preprocessor.requires_data)
def test_in_memory_preprocessor_consume_once(self):
data_iterator = iter(self.dataset)
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=False,
)
condition_iterator = preprocessor.consume_once(
"condition", components={}, data_iterator=data_iterator, cache_samples=True
)
latent_iterator = preprocessor.consume_once(
"latent", components={}, data_iterator=data_iterator, use_cached_samples=True, drop_samples=True
)
self.assertFalse(preprocessor.requires_data)
for _ in range(self.num_items):
condition_item = next(condition_iterator)
latent_item = next(latent_iterator)
self.assertIn("caption", condition_item)
self.assertIn("video", latent_item)
self.assertEqual(condition_item["caption"], "a cat ruling the world surrounded by mystical aura")
self.assertEqual(latent_item["video"].shape[-2:], (16, 16))
self.assertFalse(preprocessor.requires_data)
def test_precomputed_preprocessor_consume(self):
data_iterator = iter(self.dataset)
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=True,
)
condition_iterator = preprocessor.consume(
"condition", components={}, data_iterator=data_iterator, cache_samples=True
)
latent_iterator = preprocessor.consume(
"latent", components={}, data_iterator=data_iterator, use_cached_samples=True, drop_samples=True
)
precomputed_data_dir = os.path.join(self.save_dir.name, PRECOMPUTED_DATA_DIR)
condition_file_list = find_files(precomputed_data_dir, "condition-*")
latent_file_list = find_files(precomputed_data_dir, "latent-*")
self.assertEqual(len(condition_file_list), 3)
self.assertEqual(len(latent_file_list), 3)
self.assertFalse(preprocessor.requires_data)
for _ in range(self.num_items):
condition_item = next(condition_iterator)
latent_item = next(latent_iterator)
self.assertIn("caption", condition_item)
self.assertIn("video", latent_item)
self.assertEqual(condition_item["caption"], "a cat ruling the world surrounded by mystical aura")
self.assertEqual(latent_item["video"].shape[-2:], (16, 16))
self.assertTrue(preprocessor.requires_data)
def test_precomputed_preprocessor_consume_once(self):
data_iterator = iter(self.dataset)
preprocessor = initialize_preprocessor(
self.rank,
self.world_size,
self.num_items,
self.processor_fn,
self.save_dir.name,
enable_precomputation=True,
)
condition_iterator = preprocessor.consume_once(
"condition", components={}, data_iterator=data_iterator, cache_samples=True
)
latent_iterator = preprocessor.consume_once(
"latent", components={}, data_iterator=data_iterator, use_cached_samples=True, drop_samples=True
)
precomputed_data_dir = os.path.join(self.save_dir.name, PRECOMPUTED_DATA_DIR)
condition_file_list = find_files(precomputed_data_dir, "condition-*")
latent_file_list = find_files(precomputed_data_dir, "latent-*")
self.assertEqual(len(condition_file_list), 3)
self.assertEqual(len(latent_file_list), 3)
self.assertFalse(preprocessor.requires_data)
for _ in range(self.num_items):
condition_item = next(condition_iterator)
latent_item = next(latent_iterator)
self.assertIn("caption", condition_item)
self.assertIn("video", latent_item)
self.assertEqual(condition_item["caption"], "a cat ruling the world surrounded by mystical aura")
self.assertEqual(latent_item["video"].shape[-2:], (16, 16))
self.assertFalse(preprocessor.requires_data)
|