{"id":13499096,"url":"https://github.com/cypw/DPNs","last_synced_at":"2025-03-29T03:32:17.637Z","repository":{"id":201534725,"uuid":"95994536","full_name":"cypw/DPNs","owner":"cypw","description":"Dual Path Networks","archived":false,"fork":false,"pushed_at":"2019-03-13T07:12:27.000Z","size":187,"stargazers_count":537,"open_issues_count":16,"forks_count":150,"subscribers_count":33,"default_branch":"master","last_synced_at":"2024-10-31T17:39:09.136Z","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/cypw.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":"2017-07-02T00:41:01.000Z","updated_at":"2024-08-13T23:21:23.000Z","dependencies_parsed_at":null,"dependency_job_id":"a6008352-5f72-4448-af1e-fb2168924067","html_url":"https://github.com/cypw/DPNs","commit_stats":null,"previous_names":["cypw/dpns"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cypw%2FDPNs","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cypw%2FDPNs/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cypw%2FDPNs/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/cypw%2FDPNs/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/cypw","download_url":"https://codeload.github.com/cypw/DPNs/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":246135766,"owners_count":20729056,"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":"2024-07-31T22:00:28.550Z","updated_at":"2025-03-29T03:32:17.341Z","avatar_url":"https://github.com/cypw.png","language":"Python","funding_links":[],"categories":["Papers\u0026Codes","\u003ca name=\"Vision\"\u003e\u003c/a\u003e2. Vision"],"sub_categories":["DPN","2.1 Image Classification"],"readme":"# Dual Path Networks\nThis repository contains the code and trained models of:\n\nYunpeng Chen, Jianan Li, Huaxin Xiao, Xiaojie Jin, Shuicheng Yan, Jiashi Feng. \"Dual Path Networks\" ([NIPS17](https://arxiv.org/abs/1707.01629)).\n\n![example](fig/overview.png)\n\n- DPNs helped us won the **1st place** in Object Localization Task in [ILSVRC 2017](http://image-net.org/challenges/LSVRC/2017/index), with all competition tasks within Top 3. (Team: [NUS-Qihoo_DPNs](http://image-net.org/challenges/LSVRC/2017/results))\n\n## Implementation\n\nDPNs are implemented by [MXNet \\@92053bd](https://github.com/cypw/mxnet/tree/92053bd3e71f687b5315b8412a6ac65eb0cc32d5).\n\n### Augmentation\n| Method         |  Settings  |\n| :------------- | :--------: |\n| Random Mirror  |    True    |\n| Random Crop    |  8% - 100% |\n| Aspect Ratio   |  3/4 - 4/3 |\n| Random HSL     | [20,40,50] |\n\u003e Note: \n\u003e We did not use PCA Lighting and any other advanced augmentation methods.\n\u003e Input images are resized by bicubic interpolation.\n\n### Normalization\nThe augmented input images are substrated by mean RGB = [ 124, 117, 104 ], and then multiplied by 0.0167.\n\n### Mean-Max Pooling\nHere, we introduce a new testing technique by using Mean-Max Pooling which can further improve the performance of a well trained CNN in the testing phase without the need of any training/fine-tuining process. This testing technique is designed for the case when the testing images is larger than training crops. The idea is to first convert a trained CNN model into a [convolutional network](http://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Long_Fully_Convolutional_Networks_2015_CVPR_paper.pdf) and then insert the following Mean-Max Pooling layer (a.k.a. [Max-Avg Pooling](https://arxiv.org/abs/1509.08985)), i.e. 0.5 * (global average pooling + global max pooling), just before the final softmax layer.\n\nBased on our observations, Mean-Max Pooling consistently boost the testing accuracy. We adopted this testing strategy in both LSVRC16 and LSVRC17.\n\n\n## Results\n\n### ImageNet-1k\n\n**Single Model, Single Crop Validation Error:**\n\n\u003cdl\u003e\n\u003ctable class=\"tg\" style=\"undefined;table-layout: fixed; width: 739px\"\u003e\n\u003ccolgroup\u003e\n\u003ccol style=\"width: 103px\"\u003e\n\u003ccol style=\"width: 92px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 68px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 62px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 96px\"\u003e\n\u003c/colgroup\u003e\n  \u003ctr\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eModel\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eSize\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eGFLOPs\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e224x224\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003cbr\u003e( with mean-max pooling )\u003c/th\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-68\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e49 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e2.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e23.57\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e6.93\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e22.15\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.90\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e21.51\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.52\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-92\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e145 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e6.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e20.73\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.37\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.34\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.66\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.04\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.53\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-98\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e236 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e11.7\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e20.15\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.15\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.94\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.44\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.72\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.40\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-131\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e304 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e16.0\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.93\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.12\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.62\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.23\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.55\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.16\u003c/td\u003e\n  \u003c/tr\u003e\n\u003c/table\u003e\n\u003c/dl\u003e\n\n### ImageNet-1k (Pretrained on ImageNet-5k)\n\n**Single Model, Single Crop Validation Error:**\n\n\u003cdl\u003e\n\u003ctable class=\"tg\" style=\"undefined;table-layout: fixed; width: 739px\"\u003e\n\u003ccolgroup\u003e\n\u003ccol style=\"width: 103px\"\u003e\n\u003ccol style=\"width: 92px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 68px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 62px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 96px\"\u003e\n\u003c/colgroup\u003e\n  \u003ctr\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eModel\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eSize\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eGFLOPs\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e224x224\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003cbr\u003e( with mean-max pooling )\u003c/th\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-68\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e49 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e2.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e22.45\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e6.09\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e20.92\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.26\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e20.62\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.07\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-92\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e145 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e6.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.98\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e5.06\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.00\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.37\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.79\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.19\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-107\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e333 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.3\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e19.75\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.94\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.34\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.19\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e18.15\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e4.03\u003c/td\u003e\n  \u003c/tr\u003e\n\u003c/table\u003e\n\u003c/dl\u003e\n\n\u003eNote: DPN-107 is not well trained.\n\n\n### ImageNet-5k\n\n**Single Model, Single Crop Validation Accuracy:**\n\n\u003cdl\u003e\n\u003ctable class=\"tg\" style=\"undefined;table-layout: fixed; width: 739px\"\u003e\n\u003ccolgroup\u003e\n\u003ccol style=\"width: 103px\"\u003e\n\u003ccol style=\"width: 92px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 68px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 62px\"\u003e\n\u003ccol style=\"width: 72px\"\u003e\n\u003ccol style=\"width: 87px\"\u003e\n\u003ccol style=\"width: 96px\"\u003e\n\u003c/colgroup\u003e\n  \u003ctr\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eModel\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eSize\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" rowspan=\"2\"\u003eGFLOPs\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e224x224\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003c/th\u003e\n    \u003cth class=\"tg-baqh\" colspan=\"2\"\u003e320x320\u003cbr\u003e( with mean-max pooling )\u003c/th\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 1\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003eTop 5\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-68\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e61 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e2.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e61.27\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e85.46\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e61.54\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e85.99\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e62.35\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e86.20\u003c/td\u003e\n  \u003c/tr\u003e\n  \u003ctr\u003e\n    \u003ctd class=\"tg-baqh\"\u003eDPN-92\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e184 MB\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e6.5\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e67.31\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e89.49\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e66.84\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e89.38\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e67.42\u003c/td\u003e\n    \u003ctd class=\"tg-baqh\"\u003e89.76\u003c/td\u003e\n  \u003c/tr\u003e\n\u003c/table\u003e\n\u003c/dl\u003e\n\n\u003eNote: The higher model complexity comes from the final classifier. Models trained on ImageNet-5k learn much richer feature representation than models trained on ImageNet-1k.\n\n### Efficiency (Training)\n\nThe training speed is tested based on [MXNet \\@92053bd](https://github.com/cypw/mxnet/tree/92053bd3e71f687b5315b8412a6ac65eb0cc32d5).\n\n**Multiple Nodes (Without specific code optimization):**\n\nModel   | CUDA\u003cbr/\u003e/cuDNN | #Node | GPU Card\u003cbr/\u003e(per node) | Batch Size\u003cbr/\u003e(per GPU) | `kvstore` | GPU Mem\u003cbr/\u003e(per GPU) | Training Speed*\u003cbr/\u003e(per node)\n:-------|:------------:|:----:|:---------------------:|:----------------------:|:---------:|:---------:|:-----------:\nDPN-68  |  8.0 / 5.1   |  10  |    4 x K80 (Tesla)    |           64           |`dist_sync`|  9337 MiB | 284 img/sec\nDPN-92  |  8.0 / 5.1   |  10  |    4 x K80 (Tesla)    |           32           |`dist_sync`|  8017 MiB | 133 img/sec\nDPN-98  |  8.0 / 5.1   |  10  |    4 x K80 (Tesla)    |           32           |`dist_sync`| 11128 MiB |  85 img/sec\nDPN-131 |  8.0 / 5.1   |  10  |    4 x K80 (Tesla)    |           24           |`dist_sync`| 11448 MiB |  60 img/sec\nDPN-107 |  8.0 / 5.1   |  10  |    4 x K80 (Tesla)    |           24           |`dist_sync`| 12086 MiB |  55 img/sec\n\n\u003e \\*This is the actual training speed, which includes `data augmentation`, `forward`, `backward`, `parameter update`, `network communication`, etc. \n\u003e MXNet is awesome, we observed a linear speedup as has been shown in [link](https://github.com/dmlc/mxnet/blob/master/example/image-classification/README.md)\n\n\n## Trained Models\n\nModel    |  Size  |  Dataset  |             MXNet Model\n:--------|:------:|:---------:|:-----------------------------------:\nDPN-68   |  49 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/5iCuZ8)\nDPN-68\\* |  49 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/GZetYA)\nDPN-68   |  61 MB |ImageNet-5k|[GoogleDrive](https://goo.gl/FEbhPS)\nDPN-92   | 145 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/U4ALbg)\nDPN-92   | 138 MB |Places365-Standard|[GoogleDrive](https://goo.gl/fRq1YM)\nDPN-92\\* | 145 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/1sbov7)\nDPN-92   | 184 MB |ImageNet-5k|[GoogleDrive](https://goo.gl/H9shRv)\nDPN-98   | 236 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/kjVsLG)\nDPN-131  | 304 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/VECv1H)\nDPN-107\\*| 333 MB |ImageNet-1k|[GoogleDrive](https://goo.gl/YtokAb)\n\n\u003e\\*Pretrained on ImageNet-5k and then fine-tuned on ImageNet-1k.\n\n\n## Third-party Implementations\n\n- [Caffe Implementation](https://github.com/soeaver/caffe-model) **with trained models** by [soeaver](https://github.com/soeaver)\n- [Chainer Implementation](https://github.com/oyam/chainer-DPNs) by [oyam](https://github.com/oyam)\n- [Keras Implementation](https://github.com/titu1994/Keras-DualPathNetworks) by [titu1994](https://github.com/titu1994)\n- [MXNet Implementation](https://github.com/miraclewkf/DPN) by [miraclewkf](https://github.com/miraclewkf)\n- [PyTorch Implementation](https://github.com/oyam/pytorch-DPNs) by [oyam](https://github.com/oyam)\n- [PyTorch Implementation](https://github.com/rwightman/pytorch-dpn-pretrained) **with trained models** by [rwightman](https://github.com/rwightman)\n\n\n## Other Resources\n\nImageNet-1k Trainig/Validation List:\n- Download link: [GoogleDrive](https://goo.gl/Ne42bM)\n\nImageNet-1k category name mapping table:\n- Download link: [GoogleDrive](https://goo.gl/YTAED5)\n\nImageNet-5k Raw Images:\n- The ImageNet-5k is a subset of ImageNet10K provided by this [paper](http://vision.stanford.edu/pdf/DengBergLiFei-Fei_ECCV2010.pdf).\n- Please download the [ImageNet10K](http://www.image-net.org/download-images) and then extract the ImageNet-5k by the list below.\n\nImageNet-5k Trainig/Validation List:\n- It contains about 5k leaf categories from ImageNet10K. There is no category overlapping between our provided ImageNet-5k and the official ImageNet-1k.\n- ~~Download link: [GoogleDrive: https://goo.gl/kNZC4j]~~\n- Download link: [GoogleDrive](https://goo.gl/XViHf3)\n- Mapping Table: [GoogleDrive](https://goo.gl/vWkHYV)\n\nPlaces365-Standard Validation List \u0026 Matlab code for 10 crops testing:\n- Download link: [GoogleDrive](https://goo.gl/jQkMpr)\n\n## Citation\nIf you use DPN in your research, please cite the paper:\n```\n@article{Chen2017,\n  title={Dual Path Networks},\n  author={Yunpeng Chen, Jianan Li, Huaxin Xiao, Xiaojie Jin, Shuicheng Yan, Jiashi Feng},\n  journal={arXiv preprint arXiv:1707.01629},\n  year={2017}\n}\n```\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcypw%2FDPNs","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fcypw%2FDPNs","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcypw%2FDPNs/lists"}