Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fixed many warnings/errors & other enhancements #25

Closed
wants to merge 8 commits into from
Closed

Fixed many warnings/errors & other enhancements #25

wants to merge 8 commits into from

Conversation

Etienne66
Copy link

@Etienne66 Etienne66 commented May 21, 2021 •
edited
Loading

  • Several warnings and errors were addressed
  • Changed log.txt and config.txt files to have the time added to them so that each new run will have a new file. Especially handy for a --resume
  • n_frames_per_video increased from 100 to 200 in order to take full advantage of all of the frames in the training datasets.
  • Images are no longer saved by default during test phase of epoch.
    • There was no way to disable with a switch prior to this change
    • Can be enabled with --save_images option
  • Now gives an estimate of completion time for the current epoch and all epochs
    • 1st training session estimate does not include the 1st test session duration for the 1st epoch.
    • When using --resume --load, start time will be recalculated based on the prior elapsed times
  • PDF plots no longer created during 1st epoch due to lack of data
    • L1, HEM, & Total Loss plots are now combined in one plot instead of 3
    • PSNR plot no longer has a legend since it was blank
  • Inference will handle border situations like this
    • For a video with 5 frames [1, 2, 3, 4, 5] it will use a list of frames [3, 2, 1, 2, 3, 4, 5, 4, 3]
    • Previous handling of the same clip would produce a singe frame [3]
    • The new result will be frames [1, 2, 3, 4, 5] with all 5 frames being deblurred

@Etienne66 Etienne66 changed the title Merged latest version of correlations.py Fixed many warnings and errors and other enhancements May 23, 2021
@Etienne66 Etienne66 changed the title Fixed many warnings and errors and other enhancements Fixed many warnings/errors & other enhancements May 23, 2021
@Etienne66
Copy link
Author

Etienne66 commented May 24, 2021 •
edited
Loading

@csbhr (Haoran Bai), I want to thank you for the excellent work you have done on this paper and the code you have written. I tried your pretrained convergence model on a video I had which has a lot of motion blur due to a shaky camera and it did about as well as "Dynamic Video Deblurring using a Locally Adaptive Linear Blur Model" but much faster of course.

I did make many changes due to running on Window 10 and a newer video card but it seems to work well. I wanted to see if training it with a larger data set would improve the quality of the output. I'm running it on the below datasets combined into one. It makes for 333 videos with 31,450 frames.

After the first 2 epochs it is already at a PSNR of 27.540. I'm not sure how you defined convergence. I guess when it stops improving after a few epochs. If I have to run it for 1,000 epochs it will take more than a year. I'm guessing it will converge a long time before that but I have to admit that I do not know much about neural networks.

This is just a hobby for me. I'm a PL/SQL developer but I have some familiarity with several programming languages.

If you have any feedback about my changes I would love to hear it.

Thanks,
@Etienne66 (Steven Wilson)

@Etienne66
Copy link
Author

I've made too many changes to be merged back to this repository

@Etienne66 Etienne66 closed this Mar 10, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

when I test model on my data,happen one error Getting an error when I try to do training.
1 participant
@Etienne66