{"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/vit-comer-vision-transformer-with","title":"ViT-CoMer: Vision Transformer with Convolutional Multi-scale Feature Interaction for Dense Predictions","arxiv_id":null,"date":"2024-03-13","proceeding":"CVPR2024 2024 3","authors":["Chunlong Xia*","Xinliang Wang∗","Feng Lv∗","Xin Hao∗","Yifeng Shi†"],"abstract":"Although Vision Transformer (ViT) has achieved significant success in computer vision, it does not perform well\r\nin dense prediction tasks due to the lack of inner-patch information interaction and the limited diversity of feature\r\nscale. Most existing studies are devoted to designing visionspecific transformers to solve the above problems, which introduce additional pre-training costs. Therefore, we present\r\na plain, pre-training-free, and feature-enhanced ViT backbone with Convolutional Multi-scale feature interaction,\r\nnamed ViT-CoMer, which facilitates bidirectional interaction between CNN and transformer. Compared to the stateof-the-art, ViT-CoMer has the following advantages: (1) We\r\ninject spatial pyramid multi-receptive field convolutional\r\nfeatures into the ViT architecture, which effectively alleviates the problems of limited local information interaction\r\nand single-feature representation in ViT. (2) We propose\r\na simple and efficient CNN-Transformer bidirectional fusion interaction module that performs multi-scale fusion\r\nacross hierarchical features, which is beneficial for handling dense prediction tasks. (3) We evaluate the performance of ViT-CoMer across various dense prediction\r\ntasks, different frameworks, and multiple advanced pretraining. Notably, our ViT-CoMer-L achieves 64.3% AP on\r\nCOCO val2017 without extra training data, and 62.1%\r\nmIoU on ADE20K val, both of which are comparable to\r\nstate-of-the-art methods. We hope ViT-CoMer can serve\r\nas a new backbone for dense prediction tasks to facilitate\r\nfuture research. The code will be released at https:\r\n//github.com/Traffic-X/ViT-CoMer.","url_abs":"https://openreview.net/forum?id=srPMwpkWbR&invitationId=thecvf.com/CVPR/2024/Conference/Submission1221/-/Camera_Ready_Revision&referrer=%5BAuthor%20Console%5D(%2Fgroup%3Fid%3Dthecvf.com%2FCVPR%2F2024%2FConference%2FAuthors%23author-tasks)","url_pdf":"https://arxiv.org/pdf/2403.07392.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":"vit-comer-vision-transformer-with","repo_url":"https://github.com/Traffic-X/ViT-CoMer","is_official":0,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"vit-comer-vision-transformer-with","repo_url":"https://github.com/chenller/mmseg-extension","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"instance-segmentation","task_name":"Instance Segmentation"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"prediction","task_name":"Prediction"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"},{"method_slug":"vision-transformer","method_name":"Vision Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/instance-segmentation-on-coco-minival","task":"Instance Segmentation","dataset":"COCO minival","model":"ViT-CoMer-L (Mask RCNN, DINOv2)","rank_in_archive_order":2,"of":93,"metrics":{"mask AP":"55.9"},"uses_additional_data":false},{"leaderboard":"/sota/object-detection-on-coco-minival","task":"Object Detection","dataset":"COCO minival","model":"ViT-CoMer","rank_in_archive_order":8,"of":220,"metrics":{"Params (M)":"363","box AP":"64.3"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-segmentation-on-ade20k-val","task":"Semantic Segmentation","dataset":"ADE20K val","model":"ViT-CoMer","rank_in_archive_order":2,"of":95,"metrics":{"mIoU":"62.1"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}