pytorch>> FMNet-pytorch>> 返回
项目作者： pvnieo

项目描述 ：
A Pytorch implementation of: "Deep Functional Maps: Structured Prediction for Dense Shape Correspondence"
高级语言： Python
项目地址: git://github.com/pvnieo/FMNet-pytorch.git
创建时间： 2020-05-03T13:59:46Z
项目社区：https://github.com/pvnieo/FMNet-pytorch
开源协议：MIT License
下载


FMNet-pytorch

A pytorch implementation of: “Deep Functional Maps: Structured Prediction for Dense Shape Correspondence” [link]

Installation

This implementation runs on python >= 3.7, use pip to install dependencies:

  1. pip3 install -r requirements.txt

Download data & preprocessing

Download the desired dataset and put it in the data folder. Multiple datasets are available here.

An example with the faust-remeshed dataset is provided.

Build shot calculator:

  1. cd fmnet/utils/shot
  2. cmake .
  3. make

If you got any errors in compiling shot, please see here.

Use fmnet/preprocess.py to calculate the Laplace decomposition, geodesic distance using the Dijkstra algorithm and the shot descriptors of input shapes, data are saved in .mat format:

  1. usage: preprocess.py [-h] [-d DATAROOT] [-sd SAVE_DIR] [-ne NUM_EIGEN] [-nj NJOBS] [--nn NN]
  3. Preprocess data for FMNet training. Compute Laplacian eigen decomposition, shot features, and geodesic distance for each shape.
  5. optional arguments:
  6.   -h, --help            show this help message and exit
  7.   -d DATAROOT, --dataroot DATAROOT
  8.                         root directory of the dataset
  9.   -sd SAVE_DIR, --save-dir SAVE_DIR
  10.                         root directory to save the processed dataset
  11.   -ne NUM_EIGEN, --num-eigen NUM_EIGEN
  12.                         number of eigenvectors kept.
  13.   -nj NJOBS, --njobs NJOBS
  14.                         Number of parallel processes to use.
  15.   --nn NN               Number of Neighbor to consider when computing geodesic matrix.

NB: if the shapes have many vertices, the computation of geodesic distance will consume a lot of memory and take a lot of time.

Usage

Use the train.py script to train the FMNet network.

  1. usage: train.py [-h] [--lr LR] [--b1 B1] [--b2 B2] [-bs BATCH_SIZE] [--n-epochs N_EPOCHS] [--dim-basis DIM_BASIS] [-nv N_VERTICES] [-nb NUM_BLOCKS] [-d DATAROOT] [--save-dir SAVE_DIR] [--n-cpu N_CPU]
  2.                 [--no-cuda] [--checkpoint-interval CHECKPOINT_INTERVAL] [--log-interval LOG_INTERVAL]
  4. Launch the training of FMNet model.
  6. optional arguments:
  7.   -h, --help            show this help message and exit
  8.   --lr LR               adam: learning rate
  9.   --b1 B1               adam: decay of first order momentum of gradient
  10.   --b2 B2               adam: decay of first order momentum of gradient
  11.   -bs BATCH_SIZE, --batch-size BATCH_SIZE
  12.                         size of the batches
  13.   --n-epochs N_EPOCHS   number of epochs of training
  14.   --dim-basis DIM_BASIS
  15.                         number of eigenvectors used for representation.
  16.   -nv N_VERTICES, --n-vertices N_VERTICES
  17.                         Number of vertices used per shape
  18.   -nb NUM_BLOCKS, --num-blocks NUM_BLOCKS
  19.                         number of resnet blocks
  20.   -d DATAROOT, --dataroot DATAROOT
  21.                         root directory of the dataset
  22.   --save-dir SAVE_DIR   root directory of the dataset
  23.   --n-cpu N_CPU         number of cpu threads to use during batch generation
  24.   --no-cuda             Disable GPU computation
  25.   --checkpoint-interval CHECKPOINT_INTERVAL
  26.                         interval between model checkpoints
  27.   --log-interval LOG_INTERVAL
  28.                         interval between logging train information

Example

  1. python3 train.py -bs 4 --n-epochs 10