def _create_file_list(
min_sequence_length: int,
output_file: str,
samples_per_video: int,
source_dir_root: str,
):
file_list = FileList(
root=source_dir_root,
classes=[AVSPEECH_NAME],
min_sequence_length=min_sequence_length,
)
source_dir_root = Path(source_dir_root)
# split between train and val
videos = sorted(source_dir_root.iterdir())
train = videos[:int(len(videos) * 0.9)]
val = videos[int(len(videos) * 0.9):]
for split, split_name in [(train, TRAIN_NAME), (val, VAL_NAME)]:
for video_folder in sorted(split):
images = sorted(video_folder.glob("*.png"))
filtered_images_idx = []
# find all frames that have at least min_sequence_length-1 preceeding
# frames
sequence_start = img_name_to_int(images[0])
last_idx = sequence_start
for list_idx, image in enumerate(images):
image_idx = img_name_to_int(image)
if last_idx + 1 != image_idx:
sequence_start = image_idx
elif image_idx - sequence_start >= min_sequence_length - 1:
filtered_images_idx.append(list_idx)
last_idx = image_idx
selected_frames = select_frames(len(filtered_images_idx),
samples_per_video)
sampled_images_idx = np.asarray(
filtered_images_idx)[selected_frames]
file_list.add_data_points(
path_list=images,
target_label=AVSPEECH_NAME,
split=split_name,
sampled_images_idx=sampled_images_idx,
)
file_list.save(output_file)
logger.info(f"{output_file} created.")
return file_list
def _create_file_list(
compressions,
data_types,
min_sequence_length,
output_file,
samples_per_video,
source_dir_root,
):
file_list = FileList(
root=source_dir_root,
classes=FaceForensicsDataStructure.METHODS,
min_sequence_length=min_sequence_length,
)
# use faceforensicsdatastructure to iterate elegantly over the correct
# image folders
source_dir_data_structure = FaceForensicsDataStructure(
source_dir_root, compressions=compressions, data_types=data_types)
_min_sequence_length = _get_min_sequence_length(source_dir_data_structure)
if _min_sequence_length < samples_per_video:
logger.warning(
f"There is a sequence that is sequence that has less frames "
f"then you would like to sample: "
f"{_min_sequence_length}<{samples_per_video}")
for split, split_name in [(TRAIN, TRAIN_NAME), (VAL, VAL_NAME),
(TEST, TEST_NAME)]:
for source_sub_dir, target in zip(
source_dir_data_structure.get_subdirs(), file_list.classes):
for video_folder in sorted(source_sub_dir.iterdir()):
if video_folder.name.split("_")[0] in split:
images = sorted(video_folder.glob("*.png"))
filtered_images_idx = []
# find all frames that have at least min_sequence_length-1 preceeding
# frames
sequence_start = img_name_to_int(images[0])
last_idx = sequence_start
for list_idx, image in enumerate(images):
image_idx = img_name_to_int(image)
if last_idx + 1 != image_idx:
sequence_start = image_idx
elif image_idx - sequence_start >= min_sequence_length - 1:
filtered_images_idx.append(list_idx)
last_idx = image_idx
selected_frames = select_frames(len(filtered_images_idx),
samples_per_video)
sampled_images_idx = np.asarray(
filtered_images_idx)[selected_frames]
file_list.add_data_points(
path_list=images,
target_label=target,
split=split_name,
sampled_images_idx=sampled_images_idx,
)
file_list.save(output_file)
logger.info(f"{output_file} created.")
return file_list
#
# images = list((root_dir / "test").glob("*.jpg"))
# f.add_data_points(images, "celeba", "test", np.arange(0, len(images)))
videos = sorted(root_dir.iterdir())
train = videos[: int(len(videos) * 0.9)]
val = videos[int(len(videos) * 0.9) :]
samples_per_video = 100
for label in train:
images = sorted(label.glob("*.png"))
f.add_data_points(
images,
"avspeech",
"train",
np.rint(
np.linspace(7, len(images), min(samples_per_video, len(images))) - 1
).astype(int),
)
for label in val:
images = sorted(label.glob("*.png"))
f.add_data_points(
images,
"avspeech",
"val",
np.rint(
np.linspace(7, len(images), min(samples_per_video, len(images))) - 1
).astype(int),
)
root_dir = Path("/data/hdd/dfdc/extracted_images")
with open(root_dir.parent / "all_metadata.json", "r") as f:
all_meta_data = json.load(f)
f = FileList(str(root_dir), classes=["FAKE", "REAL"], min_sequence_length=1)
train_data_numbers = list(range(5, 50))
val_data_numbers = list(range(5))
for train_data_number in train_data_numbers:
block = root_dir / f"extracted_images_{train_data_number}"
if block.exists():
for label in block.iterdir():
images = list(label.glob("*/*.png"))
f.add_data_points(images, label.name, "train",
np.arange(0, len(images)))
for val_data_number in val_data_numbers:
block = root_dir / f"extracted_images_{val_data_number}"
if block.exists():
for label in block.iterdir():
images = list(label.glob("*/*.png"))
f.add_data_points(images, label.name, "val",
np.arange(0, len(images)))
f.add_data_points(images, label.name, "test",
np.arange(0, len(images)))
f.save("/data/ssd1/file_lists/dfdc/5_45_split.json")
for split in [TRAIN_NAME, VAL_NAME, TEST_NAME]:
data_set = FileList.get_dataset_form_file(
