search

open-mmlab / mmsegmentation

OpenMMLab Semantic Segmentation Toolbox and Benchmark.
https://mmsegmentation.readthedocs.io/en/main/
Apache License 2.0
7.69k stars 2.53k forks source link

Models compatible with the most recent version #3627

Open jacksteussie opened 2 months ago

jacksteussie commented 2 months ago

I was training models on a binary segmentation task, and I realized that a lot of the models currently don't work currently at least for binary segmentation except for the models that are more recently updated within the repo. I found this because only those ones have worked for me in coming to what I'd consider a reasonably decent prediction. I've double checked my config files and they definitely are correct. I was just wondering if this is a well-known issue, and is there a list somewhere of models that are compatible with the most recent release? Thanks! :)

lt000715 commented 2 months ago

II have the same problem as you. I have tested that fcn,deeplabv3, and unet do not work in binary classification (mIoU=0), but MAE and convnet work well. However, they may not work well after switching datasets.

I wonder if there is an official migration list, or fix this bug soon

ayushnangia commented 1 week ago

@lt000715 Can you share about how to train MAE to do segmentation on a custom dataset?