VCIP-RGBD / DFormer

[ICLR 2024] DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation
https://yinbow.github.io/Projects/DFormer/index.html
MIT License
152 stars 24 forks source link

Training and Train split #30

Open johanna27045 opened 2 months ago

johanna27045 commented 2 months ago

Hello,

I am not sure if I understand it right, but there is the validation in testing and I wanted to look deeper into the images used here. I assumed the train set would just get split and wanted to look deeper into the split. When I saw it right in the validation ion the training the eval source is used, which would mean that the validation images are not fro the test sets and additionally when really evaluating on these eval images the images would not be unseen for the network. Have I overseen something here?