KIMGEONUNG / BigColor

[ECCV 2022] Official PyTorch implementation of "BigColor"
MIT License
143 stars 22 forks source link

Black and white video colorization #7

Closed nass2020 closed 2 years ago

nass2020 commented 2 years ago

Hello, congratulations for this work, how can I colorize a black and white video with this python code? I could not get the result of a colorized image in %04d.png output,

how can I use each pre-trained model separately with %04d.png output? to use ffmpeg to get the result in video.mp4 Looking forward to your reply, please accept my best regards.

KIMGEONUNG commented 2 years ago

Thank you for your attention on our work.

To reply your question, at first, if you want to colorize a video, you have to extract all frame from video file, colorize each frame using the "colorize_real.py" script, and convert all colorized frames into a video file. There are so many example codes for video-to-frames and frames-to-video, so you can easily find that with googling.

If my answer is not what you want, please let me know.

nass2020 commented 2 years ago

Thank you for your attention on our work.

To reply your question, at first, if you want to colorize a video, you have to extract all frame from video file, colorize each frame using the "colorize_real.py" script, and convert all colorized frames into a video file. There are so many example codes for video-to-frames and frames-to-video, so you can easily find that with googling.

If my answer is not what you want, please let me know.

Hello Thank you very much for your answer, I managed to colorize all the images in the video, but the concern, that each image is colorized in 3 times with 3 different models with an output (name + '_c%03d.jpg), I would like the image this colorized with a single model and a single result with an output (%03d. jpg), so I can use this command (!ffmpeg -f image2 -framerate 24 -i /content/BigColor/results_real/%03d.jpg -c:v h264_nvenc -preset slow -qp 18 -pix_fmt yuv420p output2.mp4 ) to get the result of a colorized video, thank you again for this amazing work.

nass2020 commented 2 years ago

in this output colorized image (name + '_c%03d.jpg), I was not able to transform the images to a video.

alt text I would like the result of the colorized images to be in this forma %03d.jpg, for i use this command (!ffmpeg -f image2 -framerate 24 -i /content/BigColor/results_real/%03d.jpg -c:v h264_nvenc -preset slow -qp 18 -pix_fmt yuv420p output2.mp4 )

alt text

KIMGEONUNG commented 2 years ago

Here is a code branch for you. When you use the code of the branch, the one solution would be estimated for each input image, and the file name has no _c000 format.

If it is hard to change the code branch for you, just replace the file colorize_real.py with this https://github.com/KIMGEONUNG/BigColor/blob/nass2020/colorize_real.py

More specifically, I just edited the codes related to file name and the number of inferences. If any other issues remain, please let me know.

nass2020 commented 2 years ago

Here is a code branch for you. When you use the code of the branch, the one solution would be estimated for each input image, and the file name has no _c000 format.

If it is hard to change the code branch for you, just replace the file colorize_real.py with this https://github.com/KIMGEONUNG/BigColor/blob/nass2020/colorize_real.py

More specifically, I just edited the codes related to file name and the number of inferences. If any other issues remain, please let me know.

thank you very much for your answer, now I can have colorized %03d. jpg images to make an mp4 output, here is the result it's amazing: https://drive.google.com/file/d/1CNMRfWRHRwEV6e6CzrJCdy-xrdesEg61/view?usp=sharing

the only problem is that i couldn't get a stable color, is there any way to correct this please? if so, this would be the best colorizer i've seen so far.

Thank you very much for this phenomenal work, I look forward to your response, best regards

KIMGEONUNG commented 2 years ago

Unfortunately, our colorization method cannot sustain temporal color consistency, which is our future work. Again, thank you very much your attention.