{"id":13525858,"url":"https://github.com/bharatsingh430/soft-nms","last_synced_at":"2025-04-04T12:07:33.107Z","repository":{"id":47600926,"uuid":"88193556","full_name":"bharatsingh430/soft-nms","owner":"bharatsingh430","description":"Object Detection","archived":false,"fork":false,"pushed_at":"2019-04-04T02:47:55.000Z","size":8718,"stargazers_count":779,"open_issues_count":19,"forks_count":225,"subscribers_count":33,"default_branch":"master","last_synced_at":"2025-03-28T11:07:55.707Z","etag":null,"topics":["coco","detector","pascal-voc"],"latest_commit_sha":null,"homepage":null,"language":"Jupyter Notebook","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/bharatsingh430.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}},"created_at":"2017-04-13T18:14:46.000Z","updated_at":"2025-03-03T12:38:15.000Z","dependencies_parsed_at":"2022-09-10T09:23:35.643Z","dependency_job_id":null,"html_url":"https://github.com/bharatsingh430/soft-nms","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/bharatsingh430%2Fsoft-nms","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bharatsingh430%2Fsoft-nms/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bharatsingh430%2Fsoft-nms/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bharatsingh430%2Fsoft-nms/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/bharatsingh430","download_url":"https://codeload.github.com/bharatsingh430/soft-nms/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247174418,"owners_count":20896078,"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":["coco","detector","pascal-voc"],"created_at":"2024-08-01T06:01:22.904Z","updated_at":"2025-04-04T12:07:33.082Z","avatar_url":"https://github.com/bharatsingh430.png","language":"Jupyter Notebook","funding_links":[],"categories":["Jupyter Notebook","API"],"sub_categories":[],"readme":"# Soft-NMS\n\nThis repository includes the code for Soft-NMS. Soft-NMS is integrated with two object detectors, R-FCN and Faster-RCNN. The Soft-NMS paper can be found [here](https://arxiv.org/pdf/1704.04503.pdf).\n\nSoft-NMS is complementary to multi-scale testing and iterative bounding box regression. Check [MSRA](http://presentations.cocodataset.org/COCO17-Detect-MSRA.pdf) slides from the COCO 2017 challenge. \n\n**8 out of top 15 submissions used Soft-NMS in the [COCO 2017 detection challenge](http://cocodataset.org/#detections-leaderboard)!.**\n\nWe are also making our ICCV [reviews](http://www.cs.umd.edu/~bharat/reviews.html) and our [rebuttal](http://www.cs.umd.edu/~bharat/rebuttal.html) public. This should help to clarify some concerns which you may have.\n\nTo test the models with soft-NMS, clone the project and test your models as in standard object detection pipelines. This repository supports Faster-RCNN and R-FCN where an additional flag can be used for soft-NMS.\n\nThe flags are as follows,\n1) Standard NMS. Use flag `TEST.SOFT_NMS` 0\n2) Soft-NMS with linear weighting. Use flag `TEST.SOFT_NMS` 1 (this is the default option) \n3) Soft-NMS with Gaussian weighting. Use flag `TEST.SOFT_NMS` 2\n\nIn addition, you can specify the sigma parameter for Gaussian weighting and the threshold parameter for linear weighting. Detections below 0.001 are discarded. For integrating soft-NMS in your code, refer to `cpu_soft_nms` function in `lib/nms/cpu_nms.pyx` and `soft_nms` wrapper function in `lib/fast_rcnn/nms_wrapper.py`. You can also implement your own weighting function in this file.\n\nFor testing a model on COCO or PASCAL, use the following script\n\n```\n./tools/test_net.py --gpu ${GPU_ID} \\\n  --def models/${PT_DIR}/${NET}/rfcn_end2end/test_agnostic.prototxt \\\n  --net ${NET_FINAL} \\\n  --imdb ${TEST_IMDB} \\\n  --cfg experiments/cfgs/rfcn_end2end_ohem_${PT_DIR}.yml \\\n  --set TEST.SOFT_NMS 1 # performs soft-NMS with linear weighting\n  ${EXTRA_ARGS}\n```\n\nGPU_ID is the GPU you want to test on\n\nNET_FINAL is the caffe-model to use\n\nPT_DIR in {pascal_voc, coco} is the dataset directory\n\nDATASET in {pascal_voc, coco} is the dataset to use\n\nTEST_IMDB in {voc_0712_test,coco_2014_minival,coco_2014_test} is the test imdb\n\nTEST.SOFT_NMS in {0,1,2} is flag for different NMS algorithms. 0 is standard NMS, 1 performs soft-NMS with linear weighting and 2 performs soft-NMS with gaussian weighting\n\nPlease refer to [py-R-FCN-multiGPU](https://github.com/bharatsingh430/py-R-FCN-multiGPU/) for details about setting up object detection pipelines.\nThe Soft-NMS repository also contains code for training these detectors on multiple GPUs. **The position sensitive ROI Pooling layer is updated so that interpolation of bins is correct, like ROIAlign in Mask RCNN**. The COCO detection model for R-FCN can be found [here](https://drive.google.com/open?id=0B6T5quL13CdHX04xN1ZQX2IyMms). All other detection models used in the paper are publicly available.\n\n#### Results on MS-COCO\n\n|                   | training data       | test data          | mAP@[0.5:0.95]   | \n|-------------------|:-------------------:|:-----------------------------:|:-----:|\n|R-FCN,       NMS   | COCO 2014 train+val -minival | COCO 2015 minival     | 33.9% |\n|R-FCN,  Soft-NMS L | COCO 2014 train+val -minival | COCO 2015 minival     | 34.8% |\n|R-FCN,  Soft-NMS G | COCO 2014 train+val -minival | COCO 2015 minival     | 35.1% |\n|F-RCNN, NMS        | COCO 2014 train+val -minival | COCO 2015 test-dev    | 24.4% |\n|F-RCNN, Soft-NMS L | COCO 2014 train+val -minival | COCO 2015 test-dev    | 25.5% |\n|F-RCNN, Soft-NMS G | COCO 2014 train+val -minival | COCO 2015 test-dev    | 25.5% |\n\nR-FCN uses ResNet-101 as the backbone CNN architecture, while Faster-RCNN is based on VGG16.\n\n### Citing Soft-NMS\n\nIf you find this repository useful in your research, please consider citing:\n\n    @article{\n      Author = {Navaneeth Bodla and Bharat Singh and Rama Chellappa and Larry S. Davis},\n      Title = {Soft-NMS -- Improving Object Detection With One Line of Code},\n      Booktitle = {Proceedings of the IEEE International Conference on Computer Vision},\n      Year = {2017}\n    }\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbharatsingh430%2Fsoft-nms","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbharatsingh430%2Fsoft-nms","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbharatsingh430%2Fsoft-nms/lists"}