ewrfcas

What are updated for V2 models?

For example, COARSE.TOPKS in cvpr_ds_config.py is [16,8,8], but in loftr_ds_quadtree.py is [32,16,16]. Will it influence the inference with a pre-trained model?

> Hi! Can you confirm which version/commit of nerfstudio you're on? Sure. My system is Centos7 and the nerfacto version is 0.1.8. It seems that nerfacto could not be updated...


Thanks. I force to install nerfstudio==1.0.0 without open3d>=0.16.0. It works well now!

I think that there is no need to retrain the model. Both crop/resize can handle rectangular images. Besides, I think that our method could also work for directly inpainting rectangular...

Hi, we only resize all images into squares without cropping. And only flip augmentation is used during the training.

@shawei3000 thanks for your correction

@woshiyyya It is not a official version. I find that the earlier version of the paper can achieve better performance in the pytorch implement. Maybe I will verify the earlier...

@doris-he , this problem maybe caused by missing the spacy language package?