sheldontsui / gof_neurips2021 Goto Github PK

Could you please supplement curriculums.py with the hyperparameters about Carla's dataset? I ran conjecture experiments on the hyperparameters based on the previous three datasets, but could not train the same results as in the paper.

Hi, thanks for your work and released code! Could you share the correct outward-facing faces so that we can train from scratch?

Hi, thanks for releasing the code. I notice the script of rendering mesh only supports rotation on yaw. Could you please release scripts that also support rotation on the pitch, such as gif on your project page?

Hello, when trying to train the model by myself, I met the following error:

Traceback (most recent call last):
  File ".../site-packages/torch/multiprocessing/spawn.py", line 59, in _wrap
    fn(i, *args)
  File ".../GOF_NeurIPS2021/train.py", line 340, in train
    scaler.step(optimizer_G)
  File ".../site-packages/torch/cuda/amp/grad_scaler.py", line 337, in step
    assert len(optimizer_state["found_inf_per_device"]) > 0, "No inf checks were recorded for this optimizer."

The environment is the same as in requirements.txt (besides, the package name mcubes should be PyMCubes?).
I tried to comment that line in grad_scaler.py, although it can train now, the results seem not converging (output is still random noise after around 30000 steps).
Any help would be appreciated!

First of all, thank you for sharing this brilliant work. I discrovered a possible erorr in the Opacity Regularization formula in the paper while trying to implement this algorithm. The problem is that the opacity term described in the paper approaches negative infinity. According to the definition of entropy, I think the formula should be something like

Congrats on this great work, and thanks a lot for open-sourcing the code!
I have some trouble with the Carla dataset. Where did you get the Carla dataset？I can‘t find the pi-gan’s Carla dataset. And it seems different from Graf's. Can you provide me with a link to download the Carla dataset in your paper?
Thanks a lot in advance.

Could you please open the code of calculation of scale-invariant depth error (SIDE) and mean geodesic distance (MAD)?

Thank you very much.

Congrats on this great work, and thanks a lot for open-sourcing the code!
In the paper, you mention that “we crop all images in CelebA from the top of the hair to the bottom of the chin as a pre-processing step.” Can you provide specific steps or code scripts for pre-processing methods？
Thanks a lot in advance.