How to reconstruct environment map?

[xqwang14]

I tried both train and test code on Laval Indoor dataset, and from the test code, I get one result as the image shown above. It seems like the Guassian Map that mentioned in the paper, and I wonder how to reconstruct the environment map from this image.

Thank you so much for your excellent work and I look forward to any reply.

[fnzhan]

A generation network is followed to translate gaussian map to illumination maps, i.e., GenProjector.
The generation network is highly biased to the training scene, I recommend you to train it with your data.

[najmemhmdb]

To use GenProjector, I need latest_net_G.pth file. Could you please share its weights with me?

Thank you.

[ad45675]

hi @a978908609 ,

when i run the training code , i got some error,

Traceback (most recent call last):
File "train.py", line 97, in
dist_emloss = Sam_Loss(dist_pred, dist_gt).sum() * 1000.0
File "/venv/py37_zero-XRWy4lKA/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
TypeError: forward() missing 1 required positional argument: 'geometry'
so you have use the geometry in the training?