search

pmj110119 / RenderOcc

[ICRA 2024] RenderOcc: Vision-Centric 3D Occupancy Prediction with 2D Rendering Supervision. (Former version: UniOcc)
432 stars 24 forks source link

Reproduce results #18

Closed samernanoxx closed 11 months ago

samernanoxx commented 11 months ago

Hi, has anyone been able to reproduce the results from the paper? I am using the same config as well as the same total batch size of 16 on 4xNVIDIA RTX A6000 (samples_per_gpu=4). I have the same torch==1.10.1+cu111, torchvision==0.10.1 and mmcv-full==1.6.0 versions. The training loss reaches "0.6" at epoch 1, starts to fluctuate, and never converges. The eval mIoU is around 3. Thanks!

samernanoxx commented 11 months ago

Okay I saw similiar issue https://github.com/pmj110119/RenderOcc/issues/10 already posted. Will try to reproduce this!

pmj110119 commented 11 months ago

Have you used an earlier commit version?

There were bugs in the early version. But the latest commit should be replicable.