Implementation difference from the paper

[xinyi-spec]

Hi, thanks for sharing your inspiring work.

I found a slight difference between the implementation and the paper description.

In the paper the color_network takes the gradient on the canonical points as input normal.

However, it seems that in the code, the color_network takes the gradient on the observation point as input.

NDR-code/models/renderer.py

Lines 221 to 227 in f842e41

	gradient_o = torch.autograd.grad(
	outputs=y,
	inputs=x,
	grad_outputs=d_output,
	create_graph=True,
	retain_graph=True,
	only_inputs=True)[0]

NDR-code/models/renderer.py

Lines 265 to 266 in f842e41

	sampled_color = color_network(appearance_code, pts_canonical, gradients_o, \
	dirs_c, feature_vector, alpha_ratio).reshape(batch_size, n_samples, 3)

Any hint on this?
Thanks

[RainbowRui]

It is a good question. It takes the gradient on the observation point as input. I need to change $\textbf{p}$ to $\textbf{p}_i$ in the paper.