{"id":13520208,"url":"https://github.com/theodoruszq/PML","last_synced_at":"2025-03-31T16:30:58.245Z","repository":{"id":112584795,"uuid":"223394778","full_name":"sydney0zq/PML","owner":"sydney0zq","description":"Re-implementation of \"Blazingly Fast Video Object Segmentation with Pixel-Wise Metric Learning\"","archived":false,"fork":false,"pushed_at":"2019-11-22T12:19:48.000Z","size":28,"stargazers_count":3,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"master","last_synced_at":"2024-08-02T05:23:01.685Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/sydney0zq.png","metadata":{"files":{"readme":"README.txt","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null}},"created_at":"2019-11-22T12:04:02.000Z","updated_at":"2022-07-28T02:26:54.000Z","dependencies_parsed_at":"2023-05-16T18:00:25.332Z","dependency_job_id":null,"html_url":"https://github.com/sydney0zq/PML","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sydney0zq%2FPML","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sydney0zq%2FPML/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sydney0zq%2FPML/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sydney0zq%2FPML/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/sydney0zq","download_url":"https://codeload.github.com/sydney0zq/PML/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":222670691,"owners_count":17020513,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-08-01T05:02:14.086Z","updated_at":"2025-03-31T16:30:52.956Z","avatar_url":"https://github.com/sydney0zq.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"README.txt\n\n:Author: qiang.zhou\n:Email: theodoruszq@gmail.com\n:Date: 2018-10-11 21:46\n\n\n\nProject description:\n\n    This project dedicates to reproduce CVPR 18 paper 'Blazingly Fast Video Object \n    Segmentation with Pixel-Wise Metric Learning'. Author Chen wants to firstly\n    embed frames into an embedding space, and then use metric learning to retrieve\n    foreground and background pixels under the guide of first frame and annotation, \n    which is a novel way to do Video Object Segmentation task.\n\n    This project tries to reproduce the results reported in his paper, but finally\n    has a gap about 0.8~1.5. However, I think it is enough to do further research.\n\n\n================================================================\n\nDeep learning:\n\n    1. Data preparation\n        DAIVS\n        + trainval\n          + Annotations\n          + ImageSets\n          + JPEGImages\n        + testdev\n        + testchallenge\n        You could download them from `http://davischallenge.org`.\n\n    2. Init model preparation\n        init_models/deeplabv2_voc.pth\n        \n        Deeplab pretrained model is borrowed from \n        `https://github.com/speedinghzl/Pytorch-Deeplab`, download it by yourself.\n        Or download from: https://drive.google.com/open?id=19bHrNKQs4JzqZpoPSO5ntwMbqWQU8TIJ\n\n    3. Start to train\n        This project could train with single or multi GPU(s). You could choose one\n        depending on resources you own.\n\n        :Single GPU:\n        `CUDA_VISIBLE_DEVICES=0 python3 train.py --batch_size 4 \\\n                                                 --num_epochs 100 \\\n                                                 --learning_rate 2.5e-5 \\\n                                                 --alpha 0.7 \\\n                                                 --image_size 321 321 \\\n                                                 --gpus 0 \\\n                                                 --log_file ./experiments/run.log`\n    \n    4. Evaluate on DAVIS 16 val dataset\n        As author Chen introduces `Bilater Solver`, which is a post-process for refine\n        upsampled masks, it locates in `PROJ_ROOT/net/bs.py`，and you could run test by:\n\n        `CUDA_VISIBLE_DEVICES=0 python3 infer_bs.py`\n\n\n================================================================\n\nCoda:\n\n    Author Chen doesnot open this project's source code, therefore I could not make sure\n    my implementation absoultely right. \n\n    The accuracy report in paper:\n\n    Spat.-Temp.     Online Adapt.           Mean J          Mean F      Mean J\u0026F\n                                            72.0            73.6        72.8\n                        √                   73.2            75.0        74.1\n        √                                   74.3            78.1        76.2\n        √               √                   75.5            79.3        77.4\n\n    ---\n\n    My implemetation(Stable result):\n\n    Spat.-Temp.     Online Adapt.           Mean J          Mean F      Mean J\u0026F\n\n        √                                   73.5\n\n\nThanks:\n\n    Many thanks to https://github.com/braindeadpool/bf-vos.\n\n\n\n\n\n\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ftheodoruszq%2FPML","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Ftheodoruszq%2FPML","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Ftheodoruszq%2FPML/lists"}