-
-
Notifications
You must be signed in to change notification settings - Fork 1.8k
Description
Is there an existing issue for this?
- I have searched the existing issues
Bug description
Hello everyone,
I'm having an issue with the create_labeled_video function. Currently, I've been analyzing novel videos by setting cropping coordinates for each file separately. I do this because I labeled frames cropped around the eye and now pupil border detection performance is perfect this way (but quite finicky when using the whole frame or even with dynamic cropping).
However, when creating the labeled videos, the labels are placed according to the cropped frames label coordinates, as you can see here.
Here is a frame from a labeled video created with displaycropped=True
The issue doesn't happen with no cropping or dynamic cropping.
I found this discussion on image.sc, in which the user solved the exact same problem by using cropping=False on the config.yaml file. I checked this boolean is false for my file, so that shouldn't be an issue.
Lastly, I noticed that, when using save_frames=True, the created frame_temp folder has all the files with the correct label placing, although I haven't been able to check the finished video for this mode because the file is not written (but with no error output) and only a few frames remain in the temporary folder.
Operating System
Windows 10 (Native)
DeepLabCut version
DLC 2.2.2
DeepLabCut mode
single animal
Device type
NVIDIA GeForce RTX 2600 SUPER
Steps To Reproduce
1 - Use network trained from cropped frames
2 - Analyze novel video with cropping parameters
3 - Create labeled video
Relevant log output
No response
Anything else?
No response
Code of Conduct
- I agree to follow this project's Code of Conduct

