search

Jun-CEN / Open-world-3D-semantic-segmentation

[ECCV 2022] Open-world Semantic Segmentation for LIDAR Point Clouds
67 stars 10 forks source link

Pre-trained models and the number of epochs to reproduce results. #16

Closed Yang-Li-2000 closed 1 year ago

Yang-Li-2000 commented 1 year ago

Hi,

Could you please provide pre-trained models trained using ood_basic, ood_final, and ood_incre on the semantic kitti dataset?

We are trying to reproduce the results, but the mIoU we obtained after training is lower than expected. Specifically, at the beginning of ood_final, the valid mIoU measured by the code is 58, which is consistent with the number in the paper. However, after training using ood_final for one epoch, the mIoU dropped to 37.5. At the end of epoch 5, the mIoU is 45.

How many epochs should I train to reproduce the results in the paper (64.3) for the semantic kitti dataset?

Meanwhile, is it possible to release the checkpoints?

luoyuchenmlcv commented 1 year ago

Could you please also release the nusc checkpoints? Thanks!

Jun-CEN commented 1 year ago

Hi,

Checkpoints are available here.