Biao Zhang

I don't remember how much time needed for training. It may take two or three days. You can find my email address in `git log`.

I think what you are talking about is one-shot reconstruction which is mentioned in section 2.2 of this paper.

Thanks. You can do what ever you want with this piece of code.

我不了解TSN是什么，理论上deform conv可以替换任意的Conv2d层

Check these implementations. https://github.com/felixlaumon/deform-conv https://github.com/Zardinality/TF-deformable-conv

[These lines](https://github.com/1zb/deformable-convolution-pytorch/blob/master/src/deform_conv_cuda_kernel.cu#L190-L194) controls offsets. You can search for `offset_h` (vertical) and `offset_w` (horizontal) across all the functions of `.cu`. Let the size of input be `inC x inH x inW`,...

The forward of `ConvOffset2D` contains two part: `deformable_im2col` and `gemm`. To generalize it to 3D version, the function `deformable_im2col` needs a rewrite (`deformable_im3d2col`). Similarly, in backward propagation, following functions should...

There are two kinds of points in the reconstruction task 1) surface points (`x`) 2) query points (`points`) To reconstruct the surface, we want to be able to obtain the...

The learning of neural fields (a.k.a., neural implicit representations, coordinate-based networks) is to represent shapes with a function (an MLP in this case). 1. In testing, we are able to...

Our paper is accepted to NeurIPS 2022. All models will be released (including preprocessed data) in recent days (at least try our best). Best, Biao