search

mmmmimic / diffConvNet

PyTorch Implementation of "diffConv: Analyzing Irregular Point Clouds with an Irregular View" (ECCV'22)
MIT License
26 stars 3 forks source link

about the 3D Scene Segmentation #3

Open CodeWZT opened 1 year ago

CodeWZT commented 1 year ago

Hi, thanks for sharing the code. I try to retrain the 3d scene segmentation task on my computer. I get some bugs when I start to train the model. show as below: `Train 0, loss: 1.834570, train acc: 0.527190, train avg acc: 0.112260, train iou: 0.075878, mean iou: 0.479272

Test 0, loss: 1.757209, test acc: 0.546379, test avg acc: 0.111111, test iou: 0.068297, mean iou: 0.692278

Train 1, loss: 1.718582, train acc: 0.581674, train avg acc: 0.111090, train iou: 0.072744, mean iou: 0.69713

Test 1, loss: 1.735542, test acc: 0.546379, test avg acc: 0.111111, test iou: 0.068297, mean iou: 0.692278` As you can see, the Test mean iou is not changed, it keeps the value stable (0.692278). Can you give me some suggestions?

mmmmimic commented 1 year ago

Hi, We appreciate your interest in our work! I have rerun the whole pipeline from zero, and unfortunately, there is no such bug from my side. Could you check the preprocessed dataset as well as the dependencies? I am training the network on a GTX 1080 Ti, with PyTorch==1.9.0, and CUDA==11.6.
image

d62lu commented 1 year ago

I have the same issue when running the code, with CUDA 11.6.

mmmmimic commented 1 year ago

I have the same issue when running the code, with CUDA 11.6.

Hi! Could you provide more details for me to reproduce the bug? Thank you!

Yellowshuohahaha commented 1 year ago

I have the same issue too, with ubuntu20, pytorch1.8, and cuda11.1.

mmmmimic commented 1 year ago

Hi! My apologies for the very late reply! if you are still interested please check issue #7