{"id":13442678,"url":"https://github.com/megvii-research/RevCol","last_synced_at":"2025-03-20T15:30:27.983Z","repository":{"id":65774037,"uuid":"580996256","full_name":"megvii-research/RevCol","owner":"megvii-research","description":"Official Code of Paper \"Reversible Column Networks\" \"RevColv2\"","archived":false,"fork":false,"pushed_at":"2023-09-06T03:46:38.000Z","size":554,"stargazers_count":247,"open_issues_count":5,"forks_count":10,"subscribers_count":12,"default_branch":"main","last_synced_at":"2024-08-01T03:41:38.299Z","etag":null,"topics":["cnn","computer-vision","iclr2023","mae","pytorch","transformer","vit"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/megvii-research.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null}},"created_at":"2022-12-22T01:54:56.000Z","updated_at":"2024-07-20T16:28:09.000Z","dependencies_parsed_at":"2024-01-14T09:15:26.176Z","dependency_job_id":"d875b18a-1b46-4a80-8b86-d73efd3f4569","html_url":"https://github.com/megvii-research/RevCol","commit_stats":null,"previous_names":[],"tags_count":1,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/megvii-research%2FRevCol","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/megvii-research%2FRevCol/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/megvii-research%2FRevCol/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/megvii-research%2FRevCol/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/megvii-research","download_url":"https://codeload.github.com/megvii-research/RevCol/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":221772534,"owners_count":16878122,"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":["cnn","computer-vision","iclr2023","mae","pytorch","transformer","vit"],"created_at":"2024-07-31T03:01:49.018Z","updated_at":"2024-10-28T03:30:49.086Z","avatar_url":"https://github.com/megvii-research.png","language":"Python","funding_links":[],"categories":["Python","Summary","Fundamental MIM Methods"],"sub_categories":["MIM for Transformers and CNNs"],"readme":"# Reversible Column Networks \nThis repo is the official implementation of:\n\n### [Reversible Column Networks](https://arxiv.org/abs/2212.11696)\n[Yuxuan Cai](https://nightsnack.github.io), [Yizhuang Zhou](https://scholar.google.com/citations?user=VRSGDDEAAAAJ), [Qi Han](https://hanqer.github.io), Jianjian Sun, Xiangwen Kong, Jun Li, [Xiangyu Zhang](https://scholar.google.com/citations?user=yuB-cfoAAAAJ) \\\n[MEGVII Technology](https://en.megvii.com)\\\nInternational Conference on Learning Representations (ICLR) 2023\\\n[\\[arxiv\\]](https://arxiv.org/abs/2212.11696) \n\n### [RevColV2: Exploring Disentangled Representations in Masked Image Modeling](https://arxiv.org/abs/2309.01005)\n[Qi Han](https://hanqer.github.io), [Yuxuan Cai](https://nightsnack.github.io), [Xiangyu Zhang](https://scholar.google.com/citations?user=yuB-cfoAAAAJ) \\\n[MEGVII Technology](https://en.megvii.com)\\\n[\\[arxiv\\]](https://arxiv.org/abs/2309.01005) \n\n## Updates\n**9/06/2023***\\\nRevColv2 will be released soon!\n\n**3/15/2023***\\\nRevCol Huge checkpoint for segmentation released! Add visualization tools.\n\n**3/9/2023***\\\nDetection, Segmentation Code and Model Weights Released.\n\n***2/10/2023***\\\nRevCol model weights released.\n\n***1/21/2023***\\\nRevCol was accepted by ICLR 2023!\n\n***12/23/2022***\\\nInitial commits: codes for ImageNet-1k and ImageNet-22k classification are released.\n\n\n## To Do List\n\n\n- [x] ImageNet-1K and 22k Training Code  \n- [x] ImageNet-1K and 22k Model Weights\n- [x] Cascade Mask R-CNN COCO Object Detection Code \u0026 Model Weights\n- [x] ADE20k Semantic Segmentation Code \u0026 Model Weights\n\n\n## Introduction\nRevCol is composed of multiple copies of subnetworks, named columns respectively, between which multi-level reversible connections are employed. RevCol coud serves as a foundation model backbone for various tasks in computer vision including classification, detection and segmentation.\n\n\u003cp align=\"center\"\u003e\n\u003cimg src=\"figures/title.png\" width=100% height=100% \nclass=\"center\"\u003e\n\u003c/p\u003e\n\n## Main Results on ImageNet with Pre-trained Models\n\n| name | pretrain | resolution | #params |FLOPs | acc@1 | pretrained model | finetuned model |\n|:---------------------:| :---: | :---: | :---: | :---: | :---: | :---: | :---: |\n| RevCol-T | ImageNet-1K | 224x224 | 30M | 4.5G | 82.2 | [baidu](https://pan.baidu.com/s/1iGsbdmFcDpwviCHaajeUnA?pwd=h4tj)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_tiny_1k.pth) | - |\n| RevCol-S | ImageNet-1K | 224x224 | 60M | 9.0G | 83.5 | [baidu](https://pan.baidu.com/s/1hpHfdFrTZIPB5NTwqDMLag?pwd=mxuk)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_small_1k.pth) | - |\n| RevCol-B | ImageNet-1K | 224x224 | 138M | 16.6G | 84.1 |  [baidu](https://pan.baidu.com/s/16XIJ1n8pXPD2cXwnFX6b9w?pwd=j6x9)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_base_1k.pth) |  - |\n| RevCol-B\u003csup\u003e\\*\u003c/sup\u003e | ImageNet-22K | 224x224 | 138M | 16.6G | 85.6 |[baidu](https://pan.baidu.com/s/1l8zOFifgC8fZtBpHK2ZQHg?pwd=rh58)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_base_22k.pth)| [baidu](https://pan.baidu.com/s/1HqhDXL6OIQdn1LeM2pewYQ?pwd=1bp3)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_base_22k_1kft_224.pth)|\n| RevCol-B\u003csup\u003e\\*\u003c/sup\u003e | ImageNet-22K | 384x384 | 138M | 48.9G | 86.7 |[baidu](https://pan.baidu.com/s/1l8zOFifgC8fZtBpHK2ZQHg?pwd=rh58)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_base_22k.pth)| [baidu](https://pan.baidu.com/s/18G0zAUygKgu58s2AjCBpsw?pwd=rv86)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_base_22k_1kft_384.pth)|\n| RevCol-L\u003csup\u003e\\*\u003c/sup\u003e | ImageNet-22K | 224x224 | 273M | 39G | 86.6 |[baidu](https://pan.baidu.com/s/1ueKqh3lFAAgC-vVU34ChYA?pwd=qv5m)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_large_22k.pth)| [baidu](https://pan.baidu.com/s/1CsWmcPcwieMzXE8pVmHh7w?pwd=qd9n)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_large_22k_1kft_224.pth)|\n| RevCol-L\u003csup\u003e\\*\u003c/sup\u003e | ImageNet-22K | 384x384 | 273M | 116G | 87.6 |[baidu](https://pan.baidu.com/s/1ueKqh3lFAAgC-vVU34ChYA?pwd=qv5m)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_large_22k.pth)| [baidu](https://pan.baidu.com/s/1VmCE3W3Xw6-Lo4rWrj9Xzg?pwd=x69r)/[github](https://github.com/megvii-research/RevCol/releases/download/checkpoint/revcol_large_22k_1kft_384.pth)|\n| RevCol-H\u003csup\u003e\\*+\u003c/sup\u003e  | Megdata-168M | pretrain 224 / finetune 640 | 2.1B | 2537 | 90.0 |[huggingface](https://huggingface.co/LarryTsai/RevCol/blob/main/revcol_models/classification/revcol_huge_megdata.pth)|[huggingface](https://huggingface.co/LarryTsai/RevCol/blob/main/revcol_models/classification/revcol_huge_megdata_in1k.pth)|\n\n[+]: Note that we use a slightly different model on RevCol-H with one more branch from the bottom level to the top one. Later experiments prove that this connection is unnecessary, however, consider RevCol-H's training cost, we do not retrain it.\n## Getting Started\nPlease refer to [INSTRUCTIONS.md](INSTRUCTIONS.md) for setting up, training and evaluation details.\n\n\n## Acknowledgement\nThis repo was inspired by several open source projects. We are grateful for these excellent projects and list them as follows:\n- [timm](https://github.com/rwightman/pytorch-image-models)\n- [Swin Transformer](https://github.com/microsoft/Swin-Transformer)\n- [ConvNeXt](https://github.com/facebookresearch/ConvNeXt)\n- [beit](https://github.com/microsoft/unilm/tree/master/beit)\n\n## License\nRevCol is released under the [Apache 2.0 license](LICENSE).\n\n## Contact Us\nIf you have any questions about this repo or the original paper, please contact Yuxuan at caiyuxuan@megvii.com.\n\n\n## Citation\n```\n@inproceedings{cai2022reversible,\n  title={Reversible Column Networks},\n  author={Cai, Yuxuan and Zhou, Yizhuang and Han, Qi and Sun, Jianjian and Kong, Xiangwen and Li, Jun and Zhang, Xiangyu},\n  booktitle={International Conference on Learning Representations},\n  year={2023},\n  url={https://openreview.net/forum?id=Oc2vlWU0jFY}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmegvii-research%2FRevCol","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmegvii-research%2FRevCol","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmegvii-research%2FRevCol/lists"}