├── README.md
├── core_code.py
└── img
├── towards1.png
└── towards2.png
/README.md:
--------------------------------------------------------------------------------
1 | # Towards Better Gradient Consistency for Neural Signed Distance Functions via Level Set Alignment (CVPR 2023)
2 |
3 |
4 |
5 |
6 | This repository contains the code to reproduce the results from the paper.
7 | [Towards Better Gradient Consistency for Neural Signed Distance Functions via Level Set Alignment](http://arxiv.org/abs/2305.11601/).
8 |
9 | You can find detailed usage instructions for training your own models and using pretrained models below.
10 |
11 | If you find our code or paper useful, please consider citing
12 |
13 | @inproceedings{BaoruiTowards,
14 | title = {Towards Better Gradient Consistency for Neural Signed Distance Functions via Level Set Alignment},
15 | author = {Baorui Ma and Junsheng Zhou and Yu-Shen Liu and Zhizhong Han},
16 | booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
17 | year = {2023}
18 | }
19 |
20 | ## Coming Soon
21 |
22 | Related work
23 | ```bash
24 | Pytorch
25 | https://github.com/mabaorui/NeuralPull-Pytorch
26 | https://github.com/junshengzhou/CAP-UDF
27 | Tensorflow
28 | https://github.com/mabaorui/NeuralPull
29 | https://github.com/mabaorui/OnSurfacePrior
30 | https://github.com/mabaorui/PredictableContextPrior
31 | ```
32 |
33 |
34 | ## Demo
35 |
36 | 
37 |
38 |
39 |
40 | 
41 |
42 |
43 |
44 | ## Installation
45 | First you have to make sure that you have all dependencies in place.
46 | The simplest way to do so, is to use [anaconda](https://www.anaconda.com/).
47 |
48 | You can create an anaconda environment called `tf` using
49 | ```
50 | conda env create -f tf.yaml
51 | conda activate tf
52 | ```
53 |
54 | ## ToDo
55 | In different datasets or your own data, because of the variation in point cloud density, this ['0.25' parameter](https://github.com/mabaorui/OnSurfacePrior/blob/d53bf3a7bc88837e2974ddc1fd0700ecc2641ade/onSurPrior.py#L425) has a very strong influence on the final result, which controls the distance between the query points and the point cloud. So if you want to get better results, you should adjust this parameter. We give '0.25' here as a reference value, and this value can be used for most object-level reconstructions. For the scene dataset, we will later publish the reference values for the hyperparameter settings for the scene dataset.
56 |
--------------------------------------------------------------------------------
/core_code.py:
--------------------------------------------------------------------------------
1 | # -*- coding: utf-8 -*-
2 | """
3 | Created on Mon Jun 26 19:22:12 2023
4 |
5 | @author: Admin
6 | """
7 |
8 | # pytorch version
9 | sdf_nn_output = sdf_network(pts)
10 | sdf = sdf_nn_output[:, :1]
11 | gradients = sdf_network.gradient(pts).squeeze()
12 | gradient_norm = F.normalize(gradients, dim=-1)
13 | pts_moved = pts + gradient_norm * sdf
14 |
15 | sdf_moved = sdf_network(pts_moved)[:, :1]
16 | gradient_moved = sdf_network.gradient(pts_moved).squeeze()
17 | gradient_moved_norm = F.normalize(gradient_moved, dim=-1)
18 | consis_constraint = 1 - F.cosine_similarity(gradient_moved_norm, gradient_norm, dim=-1)
19 | weight_moved = torch.exp(-self.conf.get_float('train.sharp') * torch.abs(sdf)).reshape(-1,consis_constraint.shape[-1])
20 | consis_constraint = consis_constraint * weight_moved
21 |
22 | # tensorflow version
23 | gradient_norm = F.normalize(gradient, dim=-1)
24 | coords_moved = coords + gradient_norm * pred_sdf
25 | model_moved_input = {'coords':coords_moved}
26 | model_output_moved = model(model_moved_input)
27 | coords_moved = model_output_moved['model_in']
28 | pred_sdf_moved = model_output_moved['model_out']
29 | gradient_moved = diff_operators.gradient(pred_sdf_moved, coords_moved)
30 | gradient_moved_norm = F.normalize(gradient_moved, dim=-1)
31 | consis_constraint = 1 - F.cosine_similarity(gradient_moved_norm, gradient_norm, dim=-1)
32 | weight_moved = torch.exp(-self.conf.get_float('train.sharp') * torch.abs(sdf)).reshape(-1,consis_constraint.shape[-1])
33 | consis_constraint = consis_constraint * weight_moved
--------------------------------------------------------------------------------
/img/towards1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/mabaorui/TowardsBetterGradient/a60598f55d7be951031cb21e035d2913c709d11b/img/towards1.png
--------------------------------------------------------------------------------
/img/towards2.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/mabaorui/TowardsBetterGradient/a60598f55d7be951031cb21e035d2913c709d11b/img/towards2.png
--------------------------------------------------------------------------------