search

HuguesTHOMAS / KPConv-PyTorch

Kernel Point Convolution implemented in PyTorch
MIT License
786 stars 157 forks source link

How is mIoU calculated on the Paris-Lille-3D dataset in your paper KPConv #213

Closed xiaohe3625 closed 1 year ago

xiaohe3625 commented 1 year ago

Hi Dr.HuguesTHOMAS, Thank you for your great work！I have a question for you. How is mIoU calculated on the Paris-Lille-3D dataset in your paper KPConv? Is it Val_IoUs.txt averaging each category of IoU by this text?

HuguesTHOMAS commented 1 year ago

The validation scores are computed on the whole validation clouds in our testing function: https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/1defcd75cf7c0399704a6a9f63d3a550bfb8c1c9/utils/tester.py#L176

In particular here: https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/1defcd75cf7c0399704a6a9f63d3a550bfb8c1c9/utils/tester.py#L354

The test score are values returned by an online benchmark when we submitted the test predictions: https://npm3d.fr/paris-lille-3d

xiaohe3625 commented 1 year ago

Thank you, I see that these codes save the prediction results. https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/1defcd75cf7c0399704a6a9f63d3a550bfb8c1c9/utils/tester.py#L452-L453 What should I do when I want to evaluate a point cloud scene with labels? mIoU and Acc are calculated by real labels and predicted labels, do you provide the code?

HuguesTHOMAS commented 1 year ago

You dont have to do anything, the code already does it https://github.com/HuguesTHOMAS/KPConv-PyTorch/blob/1defcd75cf7c0399704a6a9f63d3a550bfb8c1c9/utils/tester.py#L408-L413

xiaohe3625 commented 1 year ago

Thanks for your help.