def extract(model, filepath, vid):
"""Extract features by inception_v3."""
# data loader for frames in ingle video
data_loader = get_dataloader(dataset="VideoFrame",
path=filepath,
num_frames=cfg.num_frames,
batch_size=cfg.batch_size)
# extract features by inception_v3
feats = None
for step, frames in enumerate(data_loader):
print("--> extract features [{}/{}]".format(step + 1,
len(data_loader)))
feat = model(make_variable(frames))
feats = concat_feat_var(feats, feat.data.cpu())
print("--> save feats to {}".format(
cfg.inception_v3_feats_path.format(vid)))
torch.save(feats, cfg.inception_v3_feats_path.format(vid))
print("--> delete original video file: {}".format(filepath))
os.remove(filepath)
model = make_cuda(
inception_v3(pretrained=True, transform_input=True, extract_feat=True))
model.eval()
# get vid list
video_list = os.listdir(cfg.video_root)
video_list = [
os.path.splitext(v)[0] for v in video_list
if os.path.splitext(v)[1] in cfg.video_ext
]
# extract features by inception_v3
for idx, vid in enumerate(video_list):
if os.path.exists(cfg.inception_v3_feats_path.format(vid)):
print("skip {}".format(vid))
else:
print("extract feature from {} [{}/{}]".format(
vid, idx + 1, len(video_list)))
# data loader for frames decoded from several videos
data_loader = get_dataloader(dataset="FrameImage",
path=cfg.frame_root,
batch_size=cfg.batch_size,
vid=vid)
feats = None
for step, frames in enumerate(data_loader):
print("--> step [{}/{}]".format(step + 1, len(data_loader)))
feat = model(make_variable(frames))
feats = concat_feat_var(feats, feat.data.cpu())
torch.save(feats, cfg.inception_v3_feats_path.format(vid))
inception_v3_feats = torch.load(
cfg.inception_v3_feats_path.format("total"))
else:
# get inception_v3 feats list
feats_list = os.listdir(cfg.inception_v3_feats_root)
feats_list = [
v for v in feats_list
if os.path.splitext(v)[1] in cfg.inception_v3_feats_ext
]
# load inception_v3 feats
inception_v3_feats = None
for step, feat_file in enumerate(feats_list):
print("loadingg inception_v3 from {} [{}/{}]".format(
feat_file, step + 1, len(feats_list)))
feat_path = os.path.join(cfg.inception_v3_feats_root, feat_file)
feat = torch.load(feat_path)
inception_v3_feats = concat_feat_var(inception_v3_feats, feat)
# save all feats into single file
torch.save(inception_v3_feats,
cfg.inception_v3_feats_path.format("total"))
# train PCA
X = inception_v3_feats.numpy()
pca.fit(X)
# sabe PCA params
pca.save_params(filepath=cfg.pca_model)
