search

ken2576 / vision-nerf

Official PyTorch Implementation of paper "Vision Transformer for NeRF-Based View Synthesis from a Single Input Image", WACV 2023.
MIT License
107 stars 12 forks source link

visualization #6

Closed caiyongqi closed 1 year ago

caiyongqi commented 1 year ago

Could you release the code that renders the visualization video and calculates metrics? It will save me a lot of time. Thanks!

ken2576 commented 1 year ago

You can use the script provided by PixelNeRF https://github.com/sxyu/pixel-nerf/blob/master/eval/calc_metrics.py And for rendering, eval.py and gen_real.py could be helpful.

SeaBird-Go commented 1 year ago

@ken2576 Hello, when I trained the model by the default settings on the SRN chairs dataset, I found my results are very bad when compared with your pretrained model. Could you help me why this happens?

Results generated by your pretrained model: 图片

Results generated by my trained model: 图片

ken2576 commented 1 year ago

Hi

What is the batch size you used for the training? We found that larger batch size helps with this issue. The setting for the pretrained model is 16 A100s and each with a batch size of 8. If you can't fit all of them in your memory, you could consider accumulating the gradients to achieve a larger batch. But for that you might need to implement it yourself, since I didn't include code for that.