google-research / 3d-moments Goto Github PK

Great Work! And I noticed that the enhanced version DynIBaR is also your work. I just want to ask if you have any plans to publish the code. Thanks in advance!

Really cool project! I love it!

Just one question: Is the possible to speed up the inference speed? or any suggestion? thanks!

I'm trying to run the demo with the instruction given by README. But I got the following errors. Is there a specific code change needed? Thanks!

Traceback (most recent call last):
  File "demo.py", line 248, in <module>
    render(args)
  File "demo.py", line 197, in render
    model = SpaceTimeModel(args)
  File "/home/yiqinzhao/Documents/3d-moments/model.py", line 112, in __init__
    load_scheduler=load_scheduler)
  File "/home/yiqinzhao/Documents/3d-moments/model.py", line 185, in load_from_ckpt
    self.load_model(fpath, load_opt, load_scheduler)
  File "/home/yiqinzhao/Documents/3d-moments/model.py", line 158, in load_model
    self.feature_net.load_state_dict(to_load['feature_net'])
  File "/home/yiqinzhao/miniconda3/envs/3d_moments/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1052, in load_state_dict
    self.__class__.__name__, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for ResUNet:
        size mismatch for conv1.weight: copying a param with shape torch.Size([64, 5, 7, 7]) from checkpoint, the shape in current model is torch.Size([64, 4, 7, 7]).

Hello, you mentioned in the paper that you use the combination of XVFI+3D photo inpainting as a comparison, but you repos, I did not see the code of this part @qianqianwang68

when manually access the google drive link in gdown https://drive.google.com/uc?id=1alRujVVWhqysU2xD8i5JvEvkVYinxLLD -O pretrained/in download.sh, the following problems occurs：
situation 1：Access denied

situation 2：No download button

When I manually access the google drive link in
gdown https://drive.google.com/uc?id=1EhKJE27SVc32XFjYJJpDbD4ZXO5t8dnX -O data/
the following problem occurs

could you provide another link?

Hello,

The environment file is locked with specific versions.

For RTX 3090 cudatoolkit need to be upgraded.

Regards

The vimeo-90K dataset is too large (142G).
Could you provide the dataset with some smaller zip files to download ?
Thank you !