EVALUATION.md

Evaluation - Video Object Segmentation

Prepare data

Download datasets:

python -m scripts.download_datasets

Create a directory 'datasets' for the above datasets and appropriately place each dataset to have following directory structure:

datasets/
├── vos/
│   ├── DAVIS16/ 
│   │   ├── Annotations
│   │   └── ...
│   ├── DAVIS17
│   │   ├── test-dev
│   │   │   ├── Annotations
│   │   │   └── ...
│   │   └── trainval
│   │       ├── Annotations
│   └── YouTubeVOS18
│       ├── all_frames
│       │   └── valid_all_frames
│       └── valid

Testing

python tools/eval_vos.py \
  --dataset D17 \
  --sine-weights /path/to/pt_sine/pytorch_model.bin \
  --output /path/vos/log \
  --img-size 1036 \
  --pad-size 1036 \
  --num_frame 6 \
  --memory_decay_type linear \
  --memory_decay_ratio 20 \
  --fix_first_frame
 
python inference_vos/davis2017-evaluation/evaluation_method.py \
  --davis_path datasets/vos/DAVIS17/trainval \
  --set val \
  --task semi-supervised \
  --results_path /path/vos/log