search

dongzhang89 / CONTA

Implementation for paper: Causal Intervention for Weakly-Supervised Semantic Segmentation
MIT License
185 stars 27 forks source link

how to do causal intervention #18

Closed JohnDreamer closed 3 years ago

JohnDreamer commented 3 years ago

Hi, it is a great work! But I'm a little confused about your code. As you mentioned in your paper, you have done a causal intervention by using average segmentation masks of different objects. However, I have checked out all your codes, I didn't found any code which is used to introduce average segmentation masks to do a causal intervention. And there is also no code for loading average segmentation masks. Could you please point out where you deal with this? It will be very appreciated that you answer this question!

dongzhang89 commented 3 years ago

@JohnDreamer Thanks for your attention and sorry for the confusion. Intervention is performed in a new round classification. Before this, you should do two things: 1) creating a confounder set via the segmentation mask; 2) calculating m_{t+1} for each training image via Eq.(3)

yongduosui commented 3 years ago

@dongzhang89 I think these two things is the main contribution of your paper. I hope you can release all these code, thanks.