models/model.py contains the main implementation of the MVSNet architecture, which includes :
1. differentiable homography
2. feature colume warping
3. cost volume aggregation
4. depth map generation
Each module and step has clear comments to explain the operation, which can show my own understanding of this problem.
If you would like to reproduce results, please run
bash eval2_bottle.sh
The pre-generated result files are stored in outputs_final.
If you have any questions, please feel free to ask :)