Psnr reproduction problem with pretrained model

[chobao]

Hi, I can not reproduce the psnr you@Totoro97 mentioned in the supplementary materials with your pre-trained models.

Here are the psnr from pre-trained models:

Please tell me if I miss something.

[Totoro97]

Hi, I used the pretrained models in pretrained_DTU/dtu_scan*/womask_sphere/checkpoints/ to render all images. Here are my rendered results: https://drive.google.com/file/d/1AIB6-Uv-QcJl12d4rTQBcKAGnR0H3P1G/view?usp=sharing

[chobao]

Oh, thanks for the quick reply! I know where I make the mistake. I mask the background when evaluating pnsr but the statistic in the paper is evaluated with the background. When I mask the background of the images you provided and evaluate their psnr, the results are close.

Sincerely, I also need psnr on the task of novel view synthesis with the mask. The results in the paper are evaluated without the mask.

Could you share me with the pre-trained model (held out 10% of images as the test set) or rendered results on the task of novel view synthesis mentioned in the paper?

[Totoro97]

Hi, here are the rendered results and GT images in the NVS setting: https://drive.google.com/file/d/1KyY0w1FYS1xg8CzKP5qM-cgmlZxVTCbu/view?usp=sharing

In this setting I set images indexed at [8, 13, 16, 21, 26, 31, 34, (56)] for testing and the other images for training.

I haven't yet got all the rendered results in the w/ mask setting. Rendering these images may take some time, sorry for that currently I can not manage to allocate enough GPU resources for this rendering task. Hopefully I can provide the results in the next week.

Thank you!

[chobao]

Sorry for the late reply! I was busy recently. The given images are sufficient to reproduce the results mentioned in the paper. Thanks for publishing the wonderful work and active answering.

[machengcheng2016]

Hi, can you teach me how to render the test images, please? I cannot find the evaluation codes about image rendering. @Totoro97 @1612190130