{"id":13738535,"url":"https://github.com/kevin-ssy/FishNet","last_synced_at":"2025-05-08T16:34:28.224Z","repository":{"id":75632637,"uuid":"159630488","full_name":"kevin-ssy/FishNet","owner":"kevin-ssy","description":"Implementation code of the paper: FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction, NeurIPS 2018","archived":false,"fork":false,"pushed_at":"2019-05-17T09:12:38.000Z","size":189,"stargazers_count":544,"open_issues_count":0,"forks_count":93,"subscribers_count":23,"default_branch":"master","last_synced_at":"2025-04-05T17:04:33.351Z","etag":null,"topics":["deep-learning","detection-network","image-classification","imagenet","segmentation"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":false,"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/kevin-ssy.png","metadata":{"files":{"readme":"README.md","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}},"created_at":"2018-11-29T08:10:57.000Z","updated_at":"2025-01-08T10:16:48.000Z","dependencies_parsed_at":"2023-06-07T04:30:36.461Z","dependency_job_id":null,"html_url":"https://github.com/kevin-ssy/FishNet","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/kevin-ssy%2FFishNet","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kevin-ssy%2FFishNet/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kevin-ssy%2FFishNet/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kevin-ssy%2FFishNet/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/kevin-ssy","download_url":"https://codeload.github.com/kevin-ssy/FishNet/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":253105642,"owners_count":21855070,"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":["deep-learning","detection-network","image-classification","imagenet","segmentation"],"created_at":"2024-08-03T03:02:25.842Z","updated_at":"2025-05-08T16:34:27.900Z","avatar_url":"https://github.com/kevin-ssy.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"# FishNet\n\n![ ](head_pic.jpg)\n\nThis repo holds the implementation code of the paper:\n\n[FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction](http://papers.nips.cc/paper/7356-fishnet-a-versatile-backbone-for-image-region-and-pixel-level-prediction.pdf)\n, Shuyang Sun, Jiangmiao Pang, Jianping Shi, Shuai Yi, Wanli Ouyang, NeurIPS 2018.\n\nFishNet was used as a key component\n for winning the 1st place in [COCO Detection Challenge 2018](http://cocodataset.org/#detection-leaderboard).\n \n Note that the results released here are a bit better than what we have reported in the paper.\n\n### Prerequisites\n- Python 3.6.x\n- PyTorch 0.4.0+\n\n### Data Augmentation\n\n| Method | Settings |\n| -----  | -------- |\n| Random Flip | True |\n| Random Crop | 8% ~ 100% |\n| Aspect Ratio| 3/4 ~ 4/3 |\n| Random PCA Lighting | 0.1 |\n\n**Note**: We apply weight decay to all weights and biases instead of just the weights of the convolution layers.\n\n### Training\nTo train FishNet-150 with 8 GPUs and batch size 256, simply run\n```\npython main.py --config \"cfgs/fishnet150.yaml\" IMAGENET_ROOT_PATH\n```\n\n### Models\n**Models trained without tricks**\n\n|    Model   | Params | FLOPs | Top-1  | Top-5  | Baidu Yun | Google Cloud |\n| ---------- | ------ | ----- | ------ | -----  | --------- | ------------ |\n| FishNet99  | 16.62M | 4.31G | 77.41% | 93.59% | [Download](https://pan.baidu.com/s/11U3sRod1VfbDBRbmXph6KA)| [Download](https://www.dropbox.com/s/hvojbdsad5ue7yb/fishnet99_ckpt.tar?dl=0) |\n| FishNet150 | 24.96M | 6.45G | 78.14% | 93.95% | [Download](https://pan.baidu.com/s/1uOEFsBHIdqpDLrbfCZJGUg)| [Download](https://www.dropbox.com/s/hjadcef18ln3o2v/fishnet150_ckpt.tar?dl=0)\n| FishNet201 | 44.58M | 10.58G| 78.76% | 94.39% | Available Soon | Available Soon |\n\n**Models trained with cosine lr schedule (200 epochs) and label smoothing**\n\n|    Model   | Params | FLOPs | Top-1  | Top-5  | Baidu Yun | Google Cloud |\n| ---------- | ------ | ----- | ------ | -----  | --------- | ------------ |\n| FishNet150 | 24.96M | 6.45G | 79.35% | 94.75% | [Download](https://pan.baidu.com/s/1pt31cp-xGcsRJKZAPcp4yQ) | [Download](https://www.dropbox.com/s/ajy9p6f97y45f1r/fishnet150_ckpt_welltrained.tar?dl=0) |\n| FishNet201 | 44.58M | 10.58G| 79.71% | 94.79% | [Download]() | [Download](https://www.dropbox.com/s/kvz2dmxe3fzn10m/fishnet201_ckpt_welltrain.tar?dl=0) |\n\nTo load these models, e.g. FishNet150, you need to first construct your FishNet150 structure like:\n\n```\nfrom models.network_factory import fishnet150\nmodel = fishnet150()\n```\n\nand then you can load the weights from the pre-trained checkpoint by:\n```\ncheckpoint = torch.load(model_path)  #  model_path: your checkpoint path, e.g. checkpoints/fishnet150.tar\nbest_prec1 = checkpoint['best_prec1']\nmodel.load_state_dict(checkpoint['state_dict'])\noptimizer.load_state_dict(checkpoint['optimizer'])\n```\n\nNote that you do **NOT** need to decompress the model using the ```tar``` command.\nThe model you download from the cloud could be loaded directly.\n\n### TODO:\n- [x] Update our arxiv paper.\n- [x] Release pre-train models.\n- [ ] Train the model with more training tricks.\n\n### Citation\n\nIf you find our research useful, please cite the paper:\n```\n@inproceedings{sun2018fishnet,\n  title={FishNet: A Versatile Backbone for Image, Region, and Pixel Level Prediction},\n  author={Sun, Shuyang and Pang, Jiangmiao and Shi, Jianping and Yi, Shuai and Ouyang, Wanli},\n  booktitle={Advances in Neural Information Processing Systems},\n  pages={760--770},\n  year={2018}\n}\n```\n\n### Contact\nYou can contact Shuyang Sun by sending email to kevin.sysun@gmail.com\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkevin-ssy%2FFishNet","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fkevin-ssy%2FFishNet","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkevin-ssy%2FFishNet/lists"}