This repository contains the code to reproduce the results of Deep Learning course project "How good MVSNets are at Depth Fusion?".

Based on CVPR 2020 paper:

Fast-MVSNet: Sparse-to-Dense Multi-View Stereo With Learned Propagation and Gauss-Newton Refinement

git clone https://github.com/molspace/FastMVS_experiments

pip install -r requirements.txt

• Download the preprocessed DTU training data from MVSNet and unzip it to data/dtu.

• Train the network

  python fastmvsnet/train.py --cfg configs/dtu.yaml - for original FastMVSNet

  python fastmvsnet/train1.py --cfg configs/dtu.yaml - for FastMVSNet with gt_depth directly added into the input as another dimension

  python fastmvsnet/train2.py --cfg configs/dtu.yaml - for FastMVSNet with gt_depth features extracted separately and concatenated to image features

  You can change the batch size in the configuration file according to your own pc.

• Download the rectified images from DTU benchmark and unzip it to data/dtu/Eval.

• Test with the pretrained model

  python fastmvsnet/test.py --cfg configs/dtu.yaml TEST.WEIGHT outputs/pretrained.pth

We need to apply depth fusion tools/depthfusion.py to get the complete point cloud. Please refer to MVSNet for more details.

python tools/depthfusion.py -f dtu -n flow2