Finetuning with a batch

[sheryylli]

It seems that the finetune model only takes frames from the video it is currently on, which is also the last video in a batch setting. After running a batch and finetuning, only the last video seems to have an accurate pose estimation. It looks like I may have to finetune for each video. Is there a way in batch processing that will take random frames from all videos in the batch?

[Atika-Syeda]

Hi @sheryylli can you please confirm if you used step 3 on this page? The option "Choose Yes/No to add previously refined keypoints to the training set." should let you use keypoints refined from other videos. Can you please also confirm if there's a separate .npy file saved containing previously refined data and it appears once you click refine saved data? Here's a video link showing that step: https://youtu.be/aO_kXkOuadg?t=505. You can select which .npy files to use by checking the checkboxes.

[sheryylli]

Hi @Atika-Syeda yes, there is seperate .npy file and I have followed each step. I was asking regarding finetuning as a whole batch rather than each video.

[Atika-Syeda]

Hi @sheryylli currently the GUI doesn't have an option to load multiple videos to fine tune but if you follow the steps till refining keypoints and save them then you can use them to train one model for multiple videos. Hope that helps!