{"id":13576146,"url":"https://github.com/naver-ai/rdnet","last_synced_at":"2026-04-06T01:04:06.771Z","repository":{"id":230351280,"uuid":"778900658","full_name":"naver-ai/rdnet","owner":"naver-ai","description":"[ECCV2024] Official implementation of paper, \"DenseNets Reloaded: Paradigm Shift Beyond ResNets and ViTs\".","archived":false,"fork":false,"pushed_at":"2024-08-08T05:21:14.000Z","size":15438,"stargazers_count":139,"open_issues_count":0,"forks_count":5,"subscribers_count":5,"default_branch":"main","last_synced_at":"2025-02-07T10:05:00.790Z","etag":null,"topics":["backbone","classification","convolutional-neural-networks","dense-connections","densenet","eccv2024","rdnet","revisit"],"latest_commit_sha":null,"homepage":"https://arxiv.org/abs/2403.19588","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/naver-ai.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,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2024-03-28T16:22:32.000Z","updated_at":"2025-01-11T10:39:19.000Z","dependencies_parsed_at":"2024-04-23T16:33:22.937Z","dependency_job_id":"b771b3d8-2d70-427b-ac31-fce163ad65d1","html_url":"https://github.com/naver-ai/rdnet","commit_stats":null,"previous_names":["naver-ai/rdnet"],"tags_count":1,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/naver-ai%2Frdnet","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/naver-ai%2Frdnet/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/naver-ai%2Frdnet/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/naver-ai%2Frdnet/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/naver-ai","download_url":"https://codeload.github.com/naver-ai/rdnet/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247294153,"owners_count":20915332,"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":["backbone","classification","convolutional-neural-networks","dense-connections","densenet","eccv2024","rdnet","revisit"],"created_at":"2024-08-01T15:01:07.404Z","updated_at":"2025-12-14T20:04:33.445Z","avatar_url":"https://github.com/naver-ai.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\u003ch3\u003e 🎉🎉 Our paper has been accepted at ECCV 2024! Stay tuned for more updates !! 🎉🎉 \u003c/h3\u003e\n\n\u003ch2\u003e\u003ca href=\"https://arxiv.org/abs/2403.19588\"\u003eDenseNets Reloaded: Paradigm Shift Beyond ResNets and ViTs\u003c/a\u003e\u003c/h2\u003e\n\n[Donghyun Kim](https://scholar.google.co.kr/citations?hl=en\u0026user=EBC8BMAAAAAJ)\u003csup\u003e1*\u003c/sup\u003e, [Byeongho Heo](https://sites.google.com/view/byeongho-heo/home)\u003csup\u003e2\u003c/sup\u003e, [Dongyoon Han](https://dongyoonhan.github.io/)\u003csup\u003e2*\u003c/sup\u003e\n\n\u003csup\u003e1\u003c/sup\u003e\u003ca href=\"https://www.ncloud.com/\"\u003eNAVER Cloud AI\u003c/a\u003e, \u003csup\u003e2\u003c/sup\u003e\u003ca href=\"https://naver-career.gitbook.io/en/teams/clova-cic/ai-lab\"\u003eNAVER AI Lab\u003c/a\u003e\n\u003c/div\u003e\n\n\u003cp align=\"center\"\u003e\n\u003ca href=\"https://arxiv.org/abs/2403.19588\" alt=\"arXiv\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/arXiv-2403.19588-4C35F5.svg?style=flat\" /\u003e\u003c/a\u003e\n\u003ca href=\"https://github.com/naver-ai/rdnet/blob/main/LICENSE\" alt=\"license\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/license-Apache--2.0-%23B7A800\" /\u003e\u003c/a\u003e\n\u003ca href=\"https://huggingface.co/naver-ai\" alt=\"Huggingface\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/huggingface-NAVERAILab-F58336\" /\u003e\u003c/a\u003e\n\u003c/p\u003e\n    \n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"./resources/images/rdnet_reloaded.gif\" alt=\"Densenet Reloaded\" width=\"46.5%\" height=\"100%\"\u003e\n  \u003cimg src=\"./resources/images/densenet_becomes_rdnet.gif\" alt=\"Densenet becomes RDNet\" width=\"51%\" height=\"100%\"\u003e\n\u003c/p\u003e\n\nWe revitalize **Densely Connected Convolutional Networks (DenseNets)** and reveal their untapped potential to challenge the prevalent dominance of ResNet-style architectures. Our research indicates that DenseNets were previously underestimated, primarily due to conventional design choices and training methods that underexploited their full capabilities.\n\n\n\u003cbr\u003e\n\n![tradeoff with SOTA models](resources/images/tradeoff_sota.png)\n*\u003cp align=\"center\"\u003eTradeoff with RDNet (ours) and SOTA models\u003c/p\u003e*\n\n\u003cbr\u003e\n\n![tradeoff with mainstream models](resources/images/tradeoff.png)\n*\u003cp align=\"center\"\u003eTradeoff with RDNet (ours) and mainstream models\u003c/p\u003e*\n\n### Key Highlights:\n- Pilot study (**§5.1**) reveals *concatenations' effectiveness*.\n- We have meticulously *upgraded various aspects of DenseNets* (**§3.2**) through architectural tweaks and block redesigns.\n- Our revitalized DenseNets (**RDNets**) *outperform mainstream architectures* such as Swin Transformer, ConvNeXt, and DeiT-III (**§4.1**).\n\nOur work aims to reignite interest in DenseNets by demonstrating their renewed relevance and superiority in the current architectural landscape. We encourage the community to explore and build upon our findings, paving the way for further innovative contributions in deep learning architectures.\n\n*We believe that various architectural designs that have been popular recently would be combined with dense connections successfully.*\n\n## Easy to use\nRDNet is available on `timm`. You can easily use RDNet by installing the `timm` package.\n\n```python\nimport timm\n\nmodel = timm.create_model('rdnet_large', pretrained=True)\n```\n\nFor detailed usage, please refer to the [huggingface model card](https://huggingface.co/naver-ai/rdnet_tiny.nv_in1k).\n\n## Updates\n- **(2024.07.24)**: Pip installable pacakge added.\n- **(2024.04.19)**: Initial release of the repository.\n- **(2024.03.28)**: Paper is available on [arXiv](https://arxiv.org/abs/2403.19588).\n\n## Coming Soon\n- [ ] More ImageNet-22k Pretrained Models.\n- [ ] More ImageNet-1k fine-tuned models.\n- [x] Cascade Mask R-CNN with RDNet.\n- [ ] Transfer Learning with RDNet (with cifar10, cifar100, stanford cars, ...).\n\n## RDNet for Image Classification\nFor details on object detection and instance segmentation, please refer to [detection/README.md](detection/README.md).\n\nFor details on semantic segmentation, please refer to [segmentation/README.md](segmentation/README.md).\n\n\n### Model Zoo\nWe provide the pretrained models for RDNet. You can download the pretrained models from the links below.\n\n#### ImageNet-1K (pre-)trained models\n|  Model  | IMG Size | Params | FLOPs | Top-1 | Model Card                                                         |                             url                              |\n|:-------:|:--------:|:------:|:-----:|:-----:|--------------------------------------------------------------------|:------------------------------------------------------------:|\n| RDNet-T |   224    |  22M   | 5.0G  | 82.8  | [model_card](./resources/hfhub_model_cards/rdnet_tiny.nv_in1k.md)  | [HFHub](https://huggingface.co/naver-ai/rdnet_tiny.nv_in1k)  |\n| RDNet-S |   224    |  50M   | 8.7G  | 83.7  | [model_card](./resources/hfhub_model_cards/rdnet_small.nv_in1k.md) | [HFHub](https://huggingface.co/naver-ai/rdnet_small.nv_in1k) |\n| RDNet-B |   224    |  87M   | 15.4G | 84.4  | [model_card](./resources/hfhub_model_cards/rdnet_base.nv_in1k.md)  | [HFHub](https://huggingface.co/naver-ai/rdnet_base.nv_in1k)  |\n| RDNet-L |   224    |  186M  | 34.7G | 84.8  | [model_card](./resources/hfhub_model_cards/rdnet_large.nv_in1k.md) | [HFHub](https://huggingface.co/naver-ai/rdnet_large.nv_in1k) |\n\n#### ImageNet-1K fine-tuned models\n\n|     Model     | fine-tune from | IMG Size | Params | FLOPs  | Top-1 | Model Card                                                                     |                                   url                                    |\n|:-------------:|:--------------:|:--------:|:------:|:------:|:-----:|--------------------------------------------------------------------------------|:------------------------------------------------------------------------:|\n| RDNet-L (384) |    RDNet-L     |   384    |  186M  | 101.9G | 85.8  | [model_card](./resources/hfhub_model_cards/rdnet_large.nv_in1k_ft_in1k_384.md) | [HFHub](https://huggingface.co/naver-ai/rdnet_large.nv_in1k_ft_in1k_384) |\n\n\n### Training\nWe provide the graphs of the training procedure. The graph is generated by the Weights \u0026 Biases service. You can check the graph by clicking the link below.\n\nhttps://api.wandb.ai/links/dhkim0225/822w2zsj\n\nFor training commands, please refer to the [TRAINING.md](TRAINING.md).\n\n\n### Acknowledgement\nThis repository is built using the [timm](https://github.com/huggingface/pytorch-image-models), [MMDetection](https://github.com/open-mmlab/mmdetection), and [MMSegmentation](https://github.com/open-mmlab/mmsegmentation).\n\n### Citation\n```\n@misc{kim2024densenets,\n    title={DenseNets Reloaded: Paradigm Shift Beyond ResNets and ViTs}, \n    author={Donghyun Kim and Byeongho Heo and Dongyoon Han},\n    year={2024},\n    eprint={2403.19588},\n    archivePrefix={arXiv},\n}\n```\n\n### License\n\n```\nCopyright (c) 2024-present NAVER Cloud Corp.\n\nLicensed under the Apache License, Version 2.0 (the \"License\");\nyou may not use this file except in compliance with the License.\nYou may obtain a copy of the License at\n\n    http://www.apache.org/licenses/LICENSE-2.0\n\nUnless required by applicable law or agreed to in writing, software\ndistributed under the License is distributed on an \"AS IS\" BASIS,\nWITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\nSee the License for the specific language governing permissions and\nlimitations under the License.\n```\n\n### License for Dataset\nImageNet - ImageNet Terms of access, https://image-net.org/download\n\nImages from ADE20K - ADE20K Terms of Use, https://groups.csail.mit.edu/vision/datasets/ADE20K/terms/\n\nMS COCO images dataset - Creative Commons Attribution 4.0 License, https://viso.ai/computer-vision/coco-dataset/\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fnaver-ai%2Frdnet","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fnaver-ai%2Frdnet","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fnaver-ai%2Frdnet/lists"}