How to select the 40 classes in NYUv2 to train?

laughtervv / DepthAwareCNN

Depth-aware CNN for RGB-D Segmentation, ECCV 2018

MIT License

302 stars 81 forks source link

How to select the 40 classes in NYUv2 to train? #10

Open smmdream opened 5 years ago

smmdream commented 5 years ago

@laughtervv I use the model you upload 4 days ago, and the val result color is different from the lable, they can't match. out_train_15_764_gt out_train_15_764_pred the first picture is gt, and the second is the pred I just make the label [ label > 39 ] = 0, I don't know the method is right or not, could you help me

phoebe0920 commented 4 years ago

can you tell me the pytorch version of your experiment? I met some questions when I run sh make.sh. Looking forward for your response, thank you！

JamesMcCullochDickens commented 3 years ago

They come from the code provided with Learning Rich Features : https://github.com/s-gupta/rcnn-depth

You have to download the eccv14 folder as described in the readme and use the matlab code in the benchmarkData/metadata/classMapping40 matlab file.