The poor performance of ScanNet

[kk6398]

I download and test the performance of scannet_0000. The result is as follow:

Is something wrong about scannet_0000? Looking forward your reply.

[Willyzw]

Hi, thank you for trying out our pipeline! As reported in the paper, our pipeline is capable of achieving higher PSNR performance. Could you confirm if you preprocessed the data using our provided script preprocess_scannet.py? Additionally, it’s very important to crop the border area by using the argument --cropborder 12 as done in our evaluation script run_scannet.py. This removes the black borders resulting from the undistortion process. Let me know if you need further help!

[lanlan96]

Hi, I come with the same problem. I test the 'scene0169_00' sequence. I follow the processing script and use the script run_scannet.py. The only change is that I change the args.buffer from args.buffer = min(1000, N // 10 + 150) if args.buffer < 0 else args.buffer to args.buffer = min(1000, N // 10 + 600) if args.buffer < 0 else args.buffer since it comes with errors like out of index. The results seem bad. Could you please provide some suggestions?