Benchmark

Folder Structure

/data/
├── Location01/
│   ├── state1_nat/
│   │   ├── t0/ (original photos)
│   │   ├── t1/ (renders)
│   │   └── gt/
│   ├── state1_art/
│   │   ├── t0/
│   │   ├── t1/
│   │   └── gt/
│   ├── state2_nat/
│   │   ├── t0/
│   │   ├── t1/
│   │   └── gt/
│   ├── state2_art/
│   │   ├── t0/
│   │   ├── t1/
│   │   └── gt/
│   ├── state3_nat/
│   │   ├── t0/
│   │   ├── t1/
│   │   └── gt/
│   └── state3_art/
│       ├── t0/
│       ├── t1/
│       └── gt/
├── Location02/
│   └── ...
├── Location03/
│   └── ...
...
└── Location20/
    └── ...

Build Docker Image

docker build -f Dockerfile -t <image_name> .

Run Docker Container

docker run --gpus '"device=0"' --rm -v <input_path>:/data -v <output_path>:/output -it <image_name>

Output Folder

<output_path> will contain:

/output/
├── masks/   
└── eval/

Each eval contains:

mean Precision

mean Recall

mIoU

mean F1 Score

Total time

Average time per pair