scripts/extract_crop_lips_v2.py

import sys
import time

import face_alignment

sys.path.append('..')

import multiprocessing
from multiprocessing import Pool

from tqdm.auto import tqdm
from Extractor import export_frames
from helpers import *

RUN_PARALLEL = True
PARALLEL_USE_GPU = True
NUM_WORKERS = 12

LRS2_dir = '/home/milselarch/projects/SUTD/50-035/LRS2/lrs2_v1'
videos_dir = os.path.join(LRS2_dir, 'mvlrs_v1/main')
images_dir = os.path.join(LRS2_dir, 'mvlrs_v1/main_images')

usable_sets_paths = [
    '../data/LRS2_CTC2_train.txt',
    '../data/LRS2_CTC2_val.txt',
    '../data/LRS2_CTC2_test.txt'
]

USABLE_VIDEO_GROUP_NAMES = []
for usable_sets_path in usable_sets_paths:
    USABLE_VIDEO_GROUP_NAMES.extend([
        x.strip() for x in open(usable_sets_path).readlines()
    ])

USABLE_VIDEO_GROUP_NAMES = set(USABLE_VIDEO_GROUP_NAMES)
lock = multiprocessing.Lock()
group_ids = os.listdir(videos_dir)
video_paths = []

for group_id in group_ids:
    group_dirpath = os.path.join(videos_dir, group_id)
    filenames = os.listdir(group_dirpath)

    for filename in filenames:
        if not filename.endswith('.mp4'):
            continue

        basename, _ = os.path.splitext(filename)
        video_group_name = f'{group_id}/{basename}'
        if video_group_name not in USABLE_VIDEO_GROUP_NAMES:
            continue

        video_paths.append((group_id, filename))


def process_video(
    video_path, video_images_dir, use_gpu
):
    export_frames(
        video_path, video_images_dir,
        recycle_landmarks=True, use_gpu=use_gpu,
    )
    # print('PROC', video_path)

    # print('PROCESS_END', group_id, video_filename)
    return video_path, video_images_dir


def callback(result):
    pbar.desc = str(result)
    pbar.update(1)
    # pbar.refresh()
    # shutil.rmtree(vid_temp_dir)


if RUN_PARALLEL:
    pbar = tqdm(video_paths)
    pool = Pool(processes=NUM_WORKERS)
    jobs = []

    for data_pair in video_paths:
        group_id, video_filename = data_pair
        pbar.desc = str(data_pair)

        basename, _ = os.path.splitext(video_filename)
        temp_id = f'{group_id}-{basename}'
        # vid_temp_dir = os.path.join(temp_dir, temp_id)

        video_path = os.path.join(videos_dir, group_id, video_filename)
        group_images_dir = os.path.join(images_dir, group_id)
        video_images_dir = os.path.join(group_images_dir, basename)

        if not os.path.exists(group_images_dir):
            os.mkdir(group_images_dir)
        if not os.path.exists(video_images_dir):
            os.mkdir(video_images_dir)

        job_kwargs = kwargify(
            video_path=video_path,
            video_images_dir=video_images_dir,
            use_gpu=PARALLEL_USE_GPU
        )
        job = pool.apply_async(
            process_video, kwds=job_kwargs,
            callback=callback
        )

        jobs.append(job)

    print('ALL JOBS ADDED', len(jobs))
    # Wait for all tasks to complete
    for job in jobs:
        job.wait()

    pool.close()
    pool.join()
else:
    pbar = tqdm(video_paths)

    for data_pair in pbar:
        group_id, video_filename = data_pair

        basename, _ = os.path.splitext(video_filename)
        temp_id = f'{group_id}-{basename}'
        # vid_temp_dir = os.path.join(temp_dir, temp_id)

        video_path = os.path.join(videos_dir, group_id, video_filename)
        group_images_dir = os.path.join(images_dir, group_id)
        video_images_dir = os.path.join(group_images_dir, basename)

        if not os.path.exists(group_images_dir):
            os.mkdir(group_images_dir)
        if not os.path.exists(video_images_dir):
            os.mkdir(video_images_dir)

        process_video(
            video_path=video_path,
            video_images_dir=video_images_dir,
            use_gpu=True
        )

print('TOTAL VIDEOS', len(video_paths))
# print('COMPLETE EXTRACTED VIDEOS', complete_extractions)
print('>>>')