Affiliation of Author(s): 北京航空航天大学
Journal: Measurement
Key Words: 视频物体分割，掩膜质量评估，自动质量评估
Abstract: 代码下载预使用说明： ### import VOSE-Net environments 1) download docker from: https://pan.baidu.com/s/1UiZoWoCNYaqaWWKO3Zsvug?pwd=xgeu password: xgeu 2) load docker image: docker load -i vose-net_envs.tar * If successfully loaded, you can see vose-net:latest with command "docker images". ### download VOSE-Net code and datasets 1) download code file 'VOSE-NET.zip' from: https://pan.baidu.com/s/1vBVTdIZ0_0B5rEhMJegwEw?pwd=s6gt password: s6gt 2) download dataset file 'VISA.zip' from: https://pan.baidu.com/s/1-XNLZHW6q7A96nbwa-y0vw?pwd=0nmd password: 0nmd * Extract code and dataset files to $DIR_TO_CODE and $DIR_TO_DATA. ### test VOSE-Net 1) get into docker environments for the VOSE-Net: docker run -it --gpus all --mount type=bind,source=$DIR_TO_CODE/VOSE-NET/,target=/home/VOS-QA --mount type=bind,source=$DIR_TO_DATA/VISA,target=/home/VISA vose-net:latest 2) evaluate the VOSE-Net: cd ../VOS-QA/testing python eval_vosenet_val.py ### use VOSE-Net to predict mask qualities for candidate videos 1) get into docker environments for the VOSE-Net: docker run -it --gpus all --mount type=bind,source=$DIR_TO_CODE/VOSE-NET/,target=/home/VOS-QA --mount type=bind,source=$DIR_TO_DATA/VISA,target=/home/VISA vose-net:latest 2) evaluate segmentation masks qualities for videos: python infer_score_masks_vid.py $DIR_TO_VID_FRAMES $DIR_TO_VID_FLOW $DIR_TO_MASKS e.g. python infer_score_masks_vid.py ../data/exp_vid/frames/ ../data/exp_vid/flow/ ../data/exp_vid/predicted_masks/ ### train VOSE-Net 1) get into docker environments for the VOSE-Net: docker run -it --gpus all --mount type=bind,source=$DIR_TO_CODE/VOSE-NET/,target=/home/VOS-QA --mount type=bind,source=$DIR_TO_DATA/VISA,target=/home/VISA vose-net:latest 2) train network: cd ../training python solve.py * models will be saved to /VOSE-Net/snapshot/
Indexed by: Journal paper
First-Level Discipline: Optical Engineering
Document Type: J
Translation or Not: no
Date of Publication: 2022-03-19
Included Journals: SCI
Links to published journals: https://doi.org/10.1016/j.measurement.2022.111003