{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/libra-r-cnn-towards-balanced-learning-for","title":"Libra R-CNN: Towards Balanced Learning for Object Detection","arxiv_id":"1904.02701","date":"2019-04-04","proceeding":"CVPR 2019 6","authors":["Jiangmiao Pang","Kai Chen","Jianping Shi","Huajun Feng","Wanli Ouyang","Dahua Lin"],"abstract":"Compared with model architectures, the training process, which is also\ncrucial to the success of detectors, has received relatively less attention in\nobject detection. In this work, we carefully revisit the standard training\npractice of detectors, and find that the detection performance is often limited\nby the imbalance during the training process, which generally consists in three\nlevels - sample level, feature level, and objective level. To mitigate the\nadverse effects caused thereby, we propose Libra R-CNN, a simple but effective\nframework towards balanced learning for object detection. It integrates three\nnovel components: IoU-balanced sampling, balanced feature pyramid, and balanced\nL1 loss, respectively for reducing the imbalance at sample, feature, and\nobjective level. Benefitted from the overall balanced design, Libra R-CNN\nsignificantly improves the detection performance. Without bells and whistles,\nit achieves 2.5 points and 2.0 points higher Average Precision (AP) than FPN\nFaster R-CNN and RetinaNet respectively on MSCOCO.","url_abs":"http://arxiv.org/abs/1904.02701v1","url_pdf":"http://arxiv.org/pdf/1904.02701v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/open-mmlab/mmdetection","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/OceanPang/Libra_R-CNN","is_official":0,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/CVUsers/Smart-Retail-By-Efficientdet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/hualuluu/--every-day-paper--","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/PaddlePaddle/PaddleDetection","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"paddle","reach":null},{"paper_slug":"libra-r-cnn-towards-balanced-learning-for","repo_url":"https://github.com/code-implementation1/Code4/tree/main/libra-rcnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"balanced-feature-pyramid","method_name":"Balanced Feature Pyramid"},{"method_slug":"balanced-l1-loss","method_name":"Balanced L1 Loss"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"bottleneck-residual-block","method_name":"Bottleneck Residual Block"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"embedded-gaussian-affinity","method_name":"Embedded Gaussian Affinity"},{"method_slug":"fpn","method_name":"FPN"},{"method_slug":"focal-loss","method_name":"Focal Loss"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"grouped-convolution","method_name":"Grouped Convolution"},{"method_slug":"iou-balanced-sampling","method_name":"IoU-Balanced Sampling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"libra-r-cnn","method_name":"Libra R-CNN"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"non-local-block","method_name":"Non-Local Block"},{"method_slug":"non-local-operation","method_name":"Non-Local Operation"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"resnext","method_name":"ResNeXt"},{"method_slug":"resnext-block","method_name":"ResNeXt Block"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"retinanet","method_name":"RetinaNet"},{"method_slug":"step-decay","method_name":"Step Decay"}],"datasets_introduced":[],"methods_introduced":[{"slug":"libra-r-cnn","name":"Libra R-CNN","full_name":"Libra R-CNN"}],"results":[{"leaderboard":"/sota/object-detection-on-coco-minival","task":"Object Detection","dataset":"COCO minival","model":"Libra R-CNN (ResNet-50 FPN)","rank_in_archive_order":195,"of":220,"metrics":{"AP50":"59.3","AP75":"42.0","APL":"50.5","APM":"42.1","APS":"22.9","box AP":"38.5"},"uses_additional_data":false},{"leaderboard":"/sota/object-detection-on-coco","task":"Object Detection","dataset":"COCO test-dev","model":"Libra R-CNN (ResNeXt-101-FPN)","rank_in_archive_order":164,"of":225,"metrics":{"AP50":"64","AP75":"47","APL":"54.6","APM":"45.6","APS":"25.3","box mAP":"43.0"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.02701","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}