cvlab-epfl / detecting-the-unexpected

Detecting the Unexpected via Image Resynthesis
Other
56 stars 10 forks source link

how to replace to another semantic segmentation network? #11

Closed a9311072 closed 2 years ago

a9311072 commented 3 years ago

Hello @adynathos

I would like to use this lab to re-write the new network, but I also found there are many dependencies. Could you share an example to replace the segmentation network? Thanks a lot. :)

adynathos commented 3 years ago

Hello, Thank you for the interest in my work.

If you'd like to replace the sem-seg in the joint pipeline, here are the functions to change: init_semseg tr_apply_semseg - receives an image and returns the predicted classes

The sem-seg can also be replaced in the evaluation pipline init_semseg but after you do that, it is also important to re-generate the synthesized images (run_semseg then run_gen_image)

qiuqiuzyq commented 3 years ago

Thanks for sharing, So if I want to replace the GAN and Discrepancy network, just need to change the corresponding function? init_gan
init_discrepancy
tr_apply_discrepancy

a9311072 commented 2 years ago

thanks