189GB memory usage when processing a movie (around 100 minutes long)

[alastorid]

trafficstars

Hi team! I run CodeFormer in a Windows Hyper-V VM with a Tesla M40 DDA into it. When I try a movie as the input video, CodeFormer takes 189GB of memory. The input video is a movie, "Shaolin Soccer" from DVD. The data path is DVD=>mp4=>CodeFormer As it may take several weeks to complete. I afraid that it may run out of memory in certain point. May I know if the memory usage will be doubled when saving the video?

[DanvilleC]

us ffmpg to split the movie into small period and process them one by one then join together again.

[DanvilleC]

#split video into small period: the_file = "inputs/video/movie.avi" !ffmpeg -i $the_file -c copy -map 0 -segment_time 00:00:10 -f segment -reset_timestamps 1 inputs/video/splited/movie_%03d.mp4

import os import glob ipath = "inputs/video/R" done_path = "inputs/video/Done" opath = "results/R" # scan all the jpg and png images input_v_list = sorted(glob.glob(os.path.join(ipath, '*.mp4'))) test_v_num = len(input_v_list) print (f'total video : {test_v_num}')

-------------------- start to processing ---------------------

for i, img_path in enumerate(input_v_list):

clean all the intermediate results to process the next image

if isinstance(img_path, str): img_name = os.path.basename(img_path) basename, ext = os.path.splitext(img_name) save_path = os.path.join(opath, f'{img_name}') print(f'[{i+1}/{test_v_num}] U Processing: {img_name}, I:{img_path} O:{save_path}') !python inference_codeformer.py --bg_upsampler realesrgan -w 0.5 --input_path $img_path -o $opath -s 2 !mv $img_path $done_path #finally concat all file in done_path into one.