search

zhang-tao-whu / DVIS

DVIS: Decoupled Video Instance Segmentation Framework
MIT License
127 stars 6 forks source link

Is the COCO dataset only used for training segmentation models? Do tracking datasets require separate annotations? #25

Open DongLZY opened 12 months ago

DongLZY commented 12 months ago

Hello Mr. Zhang, I would like to perform instance tracking for fruits. Is it possible to achieve this using DVIS? I have already created a dataset for image segmentation and can train instance segmentation models. Do I need to annotate the tracking training dataset additionally, or is it possible to generate it directly?

zhang-tao-whu commented 12 months ago

Yes, DVIS can handle this task. You can try training DVIS using only an image segmentation dataset, but its performance may not be optimal for complex occlusion scenarios. The best advice is to combine your fruit image segmentation dataset with an OVIS dataset for joint training.

DongLZY commented 11 months ago

Thank you for your answer. Can I use other segmener like yolact? I need to improve the processing speed of the model.

zhang-tao-whu commented 11 months ago

I think it's okay, but I haven't tried it.