search

DRosemei / RoMe

MIT License
220 stars 27 forks source link

About hyperparameter settings #11

Closed yanjh97 closed 1 year ago

yanjh97 commented 1 year ago

I have a question about the consistency between the hyperparameters in Table 2 of the paper and those in configs/local_kitti.yaml. I trained according to configs/local_kitti.yaml, however, the results in PSNR and mIoU are lower than those in the paper (PSNR 21.3 vs 23.4, mIoU 70% vs 86.7%). I'm looking forward to your reply.

DRosemei commented 1 year ago

@yanjh97 Hi，you can uncomment this line for a try.

yanjh97 commented 1 year ago

Hi, thanks for your reply. This result (PSNR 21.3, mIoU 70%) was obtained with optimizing vertice z. I find that the generated mesh is partial, is this the problem? image

DRosemei commented 1 year ago

@yanjh97 It's been a long time to do this experiment, maybe I use another block for reconstruction. I uses 12 blocks to reconstruct the whole scene and you can change 'choose_point' to reconstruct other blocks. image

yanjh97 commented 1 year ago

Okay, thanks. I will try it.