{"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/light-head-r-cnn-in-defense-of-two-stage","title":"Light-Head R-CNN: In Defense of Two-Stage Object Detector","arxiv_id":"1711.07264","date":"2017-11-20","proceeding":null,"authors":["Zeming Li","Chao Peng","Gang Yu","Xiangyu Zhang","Yangdong Deng","Jian Sun"],"abstract":"In this paper, we first investigate why typical two-stage methods are not as\nfast as single-stage, fast detectors like YOLO and SSD. We find that Faster\nR-CNN and R-FCN perform an intensive computation after or before RoI warping.\nFaster R-CNN involves two fully connected layers for RoI recognition, while\nR-FCN produces a large score maps. Thus, the speed of these networks is slow\ndue to the heavy-head design in the architecture. Even if we significantly\nreduce the base model, the computation cost cannot be largely decreased\naccordingly.\n  We propose a new two-stage detector, Light-Head R-CNN, to address the\nshortcoming in current two-stage approaches. In our design, we make the head of\nnetwork as light as possible, by using a thin feature map and a cheap R-CNN\nsubnet (pooling and single fully-connected layer). Our ResNet-101 based\nlight-head R-CNN outperforms state-of-art object detectors on COCO while\nkeeping time efficiency. More importantly, simply replacing the backbone with a\ntiny network (e.g, Xception), our Light-Head R-CNN gets 30.7 mmAP at 102 FPS on\nCOCO, significantly outperforming the single-stage, fast detectors like YOLO\nand SSD on both speed and accuracy. Code will be made publicly available.","url_abs":"http://arxiv.org/abs/1711.07264v2","url_pdf":"http://arxiv.org/pdf/1711.07264v2.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":"light-head-r-cnn-in-defense-of-two-stage","repo_url":"https://github.com/makefile/frcnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"light-head-r-cnn-in-defense-of-two-stage","repo_url":"https://github.com/princefr/Light-Head.pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"light-head-r-cnn-in-defense-of-two-stage","repo_url":"https://github.com/rickyHong/pytorch-light-head-rcnn-repl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"light-head-r-cnn-in-defense-of-two-stage","repo_url":"https://github.com/MindCode-4/code-7/tree/main/light-head","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"light-head-r-cnn-in-defense-of-two-stage","repo_url":"https://github.com/reeered/light-head","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"two","task_name":"Vocal Bursts Valence Prediction"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"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":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"non-maximum-suppression","method_name":"Non Maximum Suppression"},{"method_slug":"position-sensitive-roi-pooling","method_name":"Position-Sensitive RoI Pooling"},{"method_slug":"r-cnn","method_name":"R-CNN"},{"method_slug":"r-fcn","method_name":"R-FCN"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"speed","method_name":"SPEED"},{"method_slug":"ssd","method_name":"SSD"},{"method_slug":"svm","method_name":"SVM"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1711.07264","atlas_url":"https://app.syntology.ai/?focus=1711.07264","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}