VOS_SETUP.md

Setup-VOS

Install requirements

Download models

SiamFCpp tracker

• Tencent Weiyun, code: wg47g7 with md5sum ad240

SAT segmentor

• baidu yun, code: rcsn
• Google Drive

Compile evaluation toolkit

bash compile.sh

Set datasets

Davis

Download Davis Set soft link to dataset directory

ln -s path_to_datasets datasets/DAVIS

YoutubeVOS

Download YoutubeVOS2018 Set soft link to dataset directory

ln -s path_to_datasets datasets/youtubevos

COCO

Download COCO2017 Set soft link to dataset directory

ln -s path_to_datasets datasets/coco2017

At path_to_datasets:

datasets
└── DAVIS
    ├── Annotations
    |        ├── 480p # annotation for davis2017
    |        └── 480p_2016 # annotation for davis2016
    ├── ImageSets
    └── ...
└── coco2017
    ├── annotations
    |        ├── instances_train2017.json
    |        └── instances_val2017.json
    ├── train2017
    └── val2017
└── youtubevos
    ├── train
    └── valid

Set models

At path_to_models:

models
├── siamfcpp
    |── siamfcpp-alexnet-vot-md5_18fd31a2f94b0296c08fff9b0f9ad240.pkl
├── sat
    |── sat_res50_davis17_2518a.pkl
    |── sat_res18_davis17_7ccb1.pkl
├── resnet
    |── res-50-imagenet-fff49.pkl
    |── res-18-imagenet-7ccb1.pkl