/ReLeaPS

Illumination Planning for Generalised Photometric Stereo using Reinforcement Learning

Primary LanguagePythonGNU Affero General Public License v3.0AGPL-3.0

ReLeaPS : Reinforcement Learning-based Illumination Planning for Generalized Photometric Stereo

[GitHub] [Homepage] [Video]

Prerequisites

  • The code is tested on Linux with Python 3.10.
  • Install requirements from requirements.txt before running.

Setup

This repo contains sub-module. Clone this repo:

git clone --recurse-submodules https://github.com/jhchan0805/ReLeaPS
  • To train and evaluate on CNN-PS backbone, download the pre-trained model weight_and_model.hdf5 to data according to CNN-PS.
  • To train and evaluate on PS-FCN backbone, download the pre-trained model PS-FCN_B_S_32_normalize.pth.tar to src/PS-FCN/data/models according to PS-FCN.
  • To evaluate on DiLiGenT dataset, download DiLiGenT.zip to data according to DiLiGenT.

Training

  • Download the synthetic dataset (blobs & sculpture) for training from: datasets and place under data. image

  • Run run_train.sh.

    • make train
      • 这个 代码可以让你 使用 Makefile 里面定义好的 make train 指令, 这个的作用和 run_train.sh 是一样的
    • make benchmark
      • 这个 代码可以让你 使用 Makefile 里面定义好的 make benchmark 指令

    image

Evaluation

  • Train the models yourself or download the pre-trained models from: TBD and place under data/models.
  • Run run_benchmark.sh.

Citation

If you find our work useful for your research, please consider citing:

@InProceedings{jh2023releaps,
    author = {Chan, Junhoong and Yu, Bohan and Guo, Heng and Ren, Jieji and Lu, Zongqing and Shi, Boxin},
    title = {{ReLeaPS}: Reinforcement Learning-based Illumination Planning for Generalized Photometric Stereo},
    booktitle = {Proceedings of the International Conference on Computer Vision (ICCV)},
    month = {October},
    year = {2023},
}

Copyright (c) 2022-2023 Bohan Yu. All rights reserved.
ReLeaPS is free software licensed under GNU Affero General Public License version 3 or latter.