When using extract_outlier_frames (using the ‘manual’ option) the images that get saved in the labeled-data folder are drastically different in color compared to the original video frame. I am not sure what colormap it is but does this affect the training refinement since now the images appear different? I have tried it with different videos and got the same result. My videos are recorded on GoPro Hero 4 Black. Could there be something wrong with my installation of deeplabcut?
only one older version might have done this, as far as I know. which version of DLC are you using? either way, you should upgrade with
pip install deeplabcut==2.0.9
I am using 2.0.9 version (verified by deeplabcut.version)
Also, after some digging, it does looks like it’s converting rgb to bgr. Is that correct?
thanks for bringing this to our attention! We will fix in the next update, which will be this week (2.1!)
Just as a follow-up - should I not use the bgr images produced from the extract_oultlier_frames function then? Does it affect the training results?
If you are not using color in your experiments, you can throw them in, shouldn’t hurt - just another means of augmentation