hkchengrex / MiVOS

[CVPR 2021] Modular Interactive Video Object Segmentation: Interaction-to-Mask, Propagation and Difference-Aware Fusion. Semi-supervised VOS as well!

Home Page:https://hkchengrex.com/MiVOS/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Train a model to use interactive function to refine the pretrained model.

FJGEODEV opened this issue · comments

commented

Hi there, thanks for your great work.

I am wondering what if I want to use your template to train a model to implement this "interactive" template to refine the pretrained model, how can I do?

For example, I have the existing video and ground truth. I would like to train a model using those available data. Looks like I can use those to train a propagation model, how about fusion model and mask propagation model?

Very appreciated!

The code for training the propagation module and the interaction module is in two different repo as listed in the readme file. If you want to finetune the modules you can use them.

commented

Thanks for your response. I got that.

Another issue is that I was trying to download BL30k datasets. When using the script you provided, there is a warning message that I can't download it because too many people tried to pin the website. Then I tried to download it manually through google drive, but after few minutes, it will show "failed". I am sure my internet is good and has up to 30MB/s speed, do you know any alternative way to download? Thanks.