{"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/feature-intertwiner-for-object-detection-1","title":"Feature Intertwiner for Object Detection","arxiv_id":"1903.11851","date":"2019-03-28","proceeding":"ICLR 2019 5","authors":["Hongyang Li","Bo Dai","Shaoshuai Shi","Wanli Ouyang","Xiaogang Wang"],"abstract":"A well-trained model should classify objects with a unanimous score for every\ncategory. This requires the high-level semantic features should be as much\nalike as possible among samples. To achive this, previous works focus on\nre-designing the loss or proposing new regularization constraints. In this\npaper, we provide a new perspective. For each category, it is assumed that\nthere are two feature sets: one with reliable information and the other with\nless reliable source. We argue that the reliable set could guide the feature\nlearning of the less reliable set during training - in spirit of student\nmimicking teacher behavior and thus pushing towards a more compact class\ncentroid in the feature space. Such a scheme also benefits the reliable set\nsince samples become closer within the same category - implying that it is\neasier for the classifier to identify. We refer to this mutual learning process\nas feature intertwiner and embed it into object detection. It is well-known\nthat objects of low resolution are more difficult to detect due to the loss of\ndetailed information during network forward pass (e.g., RoI operation). We thus\nregard objects of high resolution as the reliable set and objects of low\nresolution as the less reliable set. Specifically, an intertwiner is designed\nto minimize the distribution divergence between two sets. The choice of\ngenerating an effective feature representation for the reliable set is further\ninvestigated, where we introduce the optimal transport (OT) theory into the\nframework. Samples in the less reliable set are better aligned with aid of OT\nmetric. Incorporated with such a plug-and-play intertwiner, we achieve an\nevident improvement over previous state-of-the-arts.","url_abs":"http://arxiv.org/abs/1903.11851v1","url_pdf":"http://arxiv.org/pdf/1903.11851v1.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":"feature-intertwiner-for-object-detection-1","repo_url":"https://github.com/hli2020/feature_intertwiner","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"feature-intertwiner-for-object-detection-1","repo_url":"https://github.com/qiu931110/Caffe_OHEM_SoftmaxLoss","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"object","task_name":"Object"},{"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":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"bottleneck-residual-block","method_name":"Bottleneck Residual Block"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"fpn","method_name":"FPN"},{"method_slug":"faster-r-cnn","method_name":"Faster R-CNN"},{"method_slug":"feature-intertwiner","method_name":"Feature Intertwiner"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"gradient-clipping","method_name":"Gradient Clipping"},{"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":"rpn","method_name":"RPN"},{"method_slug":"randomhorizontalflip","method_name":"Random Horizontal Flip"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"roi-align","method_name":"RoIAlign"},{"method_slug":"roipool","method_name":"RoIPool"},{"method_slug":"sgd-with-momentum","method_name":"SGD with Momentum"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"weight-decay","method_name":"Weight Decay"}],"datasets_introduced":[],"methods_introduced":[{"slug":"feature-intertwiner","name":"Feature Intertwiner","full_name":"Feature Intertwiner"}],"results":[{"leaderboard":"/sota/object-detection-on-coco","task":"Object Detection","dataset":"COCO test-dev","model":"InterNet (ResNet-101-FPN, multi-scale)","rank_in_archive_order":149,"of":225,"metrics":{"AP50":"67.5","AP75":"51.1","APL":"57.7","APM":"50.3","APS":"27.2","box mAP":"44.2"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}