Hi! Thanks for open sourcing your awesome work. I have tried out your 3d editing examples in your readme:
ns-train pds --data data/yuseung \
--load-dir outputs/yuseung/nerfacto/2024-09-13_215316/nerfstudio_models/ \
--pipeline.pds.src_prompt "a photo of a sks man" \
--pipeline.pds.tgt_prompt "a photo of a sks SpiderMan" \
--pipeline.pds.sd_pretrained_model_or_path data/yuseung_dreambooth \
--machine.num-devices 1
Before this, I had downloaded the dreambooth ckpt and initialized the NeRF Scene. However, I got into two problems:
The training speed is very slow (~6 hours). My device is Nvidia A100 80G, it should be very fast (actually NeRF training completes within 20 mins). And I also tried Gaussian Splatting, it was also very slow, nearly the same with NeRF. I think something must go wrong with my trial.
The results are weird. It was expected to be similar with the result shown in your paper. However I got two weird results (Both NeRF and GS):
GS edited with pdsplat
NeRF edited with pds
Could you please help me to resolve these issues? Many Thanks!
Hi! Thanks for open sourcing your awesome work. I have tried out your 3d editing examples in your readme: ns-train pds --data data/yuseung \ --load-dir outputs/yuseung/nerfacto/2024-09-13_215316/nerfstudio_models/ \ --pipeline.pds.src_prompt "a photo of a sks man" \ --pipeline.pds.tgt_prompt "a photo of a sks SpiderMan" \ --pipeline.pds.sd_pretrained_model_or_path data/yuseung_dreambooth \ --machine.num-devices 1
Before this, I had downloaded the dreambooth ckpt and initialized the NeRF Scene. However, I got into two problems:
Could you please help me to resolve these issues? Many Thanks!