pytorch>> Pose-refinement>> 返回
项目作者： densechen

项目描述 ：
Pose refinement with differentiable rendering
高级语言： Python
项目地址: git://github.com/densechen/Pose-refinement.git
创建时间： 2020-08-12T13:36:38Z
项目社区：https://github.com/densechen/Pose-refinement
开源协议：MIT License
下载


Pose refinement with differentiable rendering

Main Idea

We follow the main idea from [1] expect using differentiable renderer instead of original pre-render step. By differentiable renderer, we can chain each step to make a global refinement.
framework
You can refer to original paper for more details.

Install

This project is based on PyTorch and pytorch3d. 

  1. conda create -n pytorch3d python=3.8
  3. conda activate pytorch3d
  4. pip install -r requirement.txt

Dataset

Please download the YCBDataset from here, and modified the dataroot in ‘settings/ycb.yaml:DATA_ROOT=YOUR DATA PATH’.

Train

Run python tools/train.py to start a new train.

Evaluation

We obey the evaluation step of DenseFusion.
Run python test.py to generate the result, which can directly use for evaluation. The result will be saved under ./result, and has the same data structure with DenseFusion.

Pretrained Model

We also provide a pretrained model, you can download them from
(links: https://pan.baidu.com/s/1Wz_3A5fzDbT8Phc1QnGobw passwd: igh7).

Visualization of Refinement Result



Reference

[1] Li Y, Wang G, Ji X, et al. Deepim: Deep iterative matching for 6d pose estimation[C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 683-698.