CVPR 2021 目标检测论文大盘点（65篇论文）

前言

一共搜集了65篇2D目标检测论文，涉及：通用目标检测、旋转目标检测、Few-shot/自监督/半监督/无监督目标检测等方向。

作者：Amusi | 来源：CVer

CVer 正式盘点CVPR 2021上各个方向的工作，本篇是热度依然很高的2D目标检测论文大盘点，之前已分享：

最新！CVPR 2021 视觉Transformer论文大盘点（43篇）
最新！CVPR 2021 OCR领域论文大盘点（22篇）

关于更多CVPR 2021的论文和开源代码，可见下面链接：

CVPR2021 Papers with Code
​

CVPR 2021 2D目标检测论文（65篇）

Amusi 一共搜集了65篇2D目标检测论文，涉及：通用目标检测、旋转目标检测、Few-shot/自监督/半监督/无监督目标检测等方向。

注意：

• 这应该是目前各平台上最新最全面的CVPR 2021 2D目标检测- 盘点资料，欢迎点赞收藏和分享
• 3D目标检测、人脸检测、异常检测等检测方向并不在本文范畴，后续将单独分享，敬请期待！
• 65篇中有超过50+篇论文都来自华人，而且至少50+篇都来自中国地区（高校、企业），其中高校以清华、中科院、国科大等为主，企业以旷视、商汤等为主。

2D目标检测

1. Scaled-YOLOv4: Scaling Cross Stage Partial Network

• 作者单位: 中央研究院, 英特尔, 静宜大学
• Paper: https://arxiv.org/abs/2011.08036
• Code: https://github.com/WongKinYiu/ScaledYOLOv4
• 中文解读: YOLOv4官方改进版来了！55.8% AP！速度最高达1774 FPS，Scaled-YOLOv4正式开源！

2. You Only Look One-level Feature

• 作者单位: 中科院, 国科大, 旷视科技
• Paper: https://arxiv.org/abs/2103.09460
• Code: https://github.com/megvii-model/YOLOF
• 中文解读: CVPR 2021 | 没有FPN！中科院&旷视提出YOLOF：你只需看一层特征

3. Sparse R-CNN: End-to-End Object Detection with Learnable Proposals

• 作者单位: 香港大学, 同济大学, 字节跳动AI Lab, 加利福尼亚大学伯克利分校
• Paper: https://arxiv.org/abs/2011.12450
• Code: https://github.com/PeizeSun/SparseR-CNN
• 中文解读: 目标检测新范式！港大同济伯克利提出Sparse R-CNN，代码刚刚开源！

4. End-to-End Object Detection with Fully Convolutional Network

• 作者单位: 旷视科技, 西安交通大学
• Paper: https://arxiv.org/abs/2012.03544
• Code: https://github.com/Megvii-BaseDetection/DeFCN

5. Dynamic Head: Unifying Object Detection Heads with Attentions

• 作者单位: 微软
• Paper: https://arxiv.org/abs/2106.08322
• Code: https://github.com/microsoft/DynamicHead
• 中文解读: 60.6 AP！打破COCO记录！微软提出DyHead：将注意力与目标检测Heads统一

6. Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection

• 作者单位: 南京理工大学, Momenta, 南京大学, 清华大学
• Paper: https://arxiv.org/abs/2011.12885
• Code: https://github.com/implus/GFocalV2
• 中文解读：CVPR 2021 | GFLV2：目标检测良心技术，无Cost涨点！

7. UP-DETR: Unsupervised Pre-training for Object Detection with Transformers

• 作者单位: 华南理工大学, 腾讯微信AI
• Paper(Oral): https://arxiv.org/abs/2011.09094
• Code: https://github.com/dddzg/up-detr
• 中文解读: CVPR 2021 Oral | Transformer再发力！华南理工和微信提出UP-DETR：无监督预训练检测器

8. MobileDets: Searching for Object Detection Architectures for Mobile Accelerators

• 作者单位: 威斯康星大学, 谷歌

• Paper: https://openaccess.thecvf.com/content/CVPR2021/papers/Xiong_MobileDets_Searching_for_Object_Detection_Architectures_for_Mobile_Accelerators_CVPR_2021_paper.pdf

• Code: https://github.com/tensorflow/models/tree/master/research/object_detection

9. Tracking Pedestrian Heads in Dense Crowd

• 作者单位: 雷恩第一大学
• Homepage: https://project.inria.fr/crowdscience/project/dense-crowd-head-tracking/
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Sundararaman_Tracking_Pedestrian_Heads_in_Dense_Crowd_CVPR_2021_paper.html
• Code1: https://github.com/Sentient07/HeadHunter
• Code2: https://github.com/Sentient07/HeadHunter%E2%80%93T
• Dataset: https://project.inria.fr/crowdscience/project/dense-crowd-head-tracking/

10. Joint-DetNAS: Upgrade Your Detector with NAS, Pruning and Dynamic Distillation

• 作者单位: 香港科技大学, 华为诺亚
• Paper: https://arxiv.org/abs/2105.12971
• Code: None

11. PSRR-MaxpoolNMS: Pyramid Shifted MaxpoolNMS with Relationship Recovery

• 作者单位: A*star, 四川大学, 南洋理工大学
• Paper: https://arxiv.org/abs/2105.12990
• Code: None

12. IQDet: Instance-wise Quality Distribution Sampling for Object Detection

• 作者单位: 旷视科技
• Paper: https://arxiv.org/abs/2104.06936
• Code: None

13. Multi-Scale Aligned Distillation for Low-Resolution Detection

• 作者单位: 香港中文大学, Adobe研究院, 思谋科技
• Paper: https://jiaya.me/papers/ms_align_distill_cvpr21.pdf
• Code: https://github.com/Jia-Research-Lab/MSAD

14. Adaptive Class Suppression Loss for Long-Tail Object Detection

• 作者单位: 中科院, 国科大, ObjectEye, 北京大学, 鹏城实验室, Nexwise

• Paper: https://arxiv.org/abs/2104.00885

• Code: https://github.com/CASIA-IVA-Lab/ACSL

15. VarifocalNet: An IoU-aware Dense Object Detector

• 作者单位: 昆士兰科技大学, 昆士兰大学
• Paper(Oral): https://arxiv.org/abs/2008.13367
• Code: https://github.com/hyz-xmaster/VarifocalNet

16. OTA: Optimal Transport Assignment for Object Detection

• 作者单位: 早稻田大学, 旷视科技

• Paper: https://arxiv.org/abs/2103.14259

• Code: https://github.com/Megvii-BaseDetection/OTA

17. Distilling Object Detectors via Decoupled Features

• 作者单位: 华为诺亚, 悉尼大学
• Paper: https://arxiv.org/abs/2103.14475
• Code: https://github.com/ggjy/DeFeat.pytorch

18. Robust and Accurate Object Detection via Adversarial Learning

• 作者单位: 谷歌, UCLA, UCSC

• Paper: https://arxiv.org/abs/2103.13886

• Code: None

19. OPANAS: One-Shot Path Aggregation Network Architecture Search for Object Detection

• 作者单位: 北京大学, Anyvision, 石溪大学
• Paper: https://arxiv.org/abs/2103.04507
• Code: https://github.com/VDIGPKU/OPANAS

20. Multiple Instance Active Learning for Object Detection

• 作者单位: 国科大, 华为诺亚, 清华大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/papers/Yuan_Multiple_Instance_Active_Learning_for_Object_Detection_CVPR_2021_paper.pdf
• Code: https://github.com/yuantn/MI-AOD

21. Towards Open World Object Detection

• 作者单位: 印度理工学院, MBZUAI, 澳大利亚国立大学, 林雪平大学
• Paper(Oral): https://arxiv.org/abs/2103.02603
• Code: https://github.com/JosephKJ/OWOD

22. RankDetNet: Delving Into Ranking Constraints for Object Detection

• 作者单位: 赛灵思
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Liu_RankDetNet_Delving_Into_Ranking_Constraints_for_Object_Detection_CVPR_2021_paper.html
• Code: None

旋转目标检测

23. Dense Label Encoding for Boundary Discontinuity Free Rotation Detection

• 作者单位: 上海交通大学, 国科大
• Paper: https://arxiv.org/abs/2011.09670
• Code1: https://github.com/Thinklab-SJTU/DCL_RetinaNet_Tensorflow
• Code2: https://github.com/yangxue0827/RotationDetection

24. ReDet: A Rotation-equivariant Detector for Aerial Object Detection

• 作者单位: 武汉大学

• Paper: https://arxiv.org/abs/2103.07733

• Code: https://github.com/csuhan/ReDet

25. Beyond Bounding-Box: Convex-Hull Feature Adaptation for Oriented and Densely Packed Object Detection

• 作者单位: 国科大, 清华大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Guo_Beyond_Bounding-Box_Convex-Hull_Feature_Adaptation_for_Oriented_and_Densely_Packed_CVPR_2021_paper.html
• Code: https://github.com/SDL-GuoZonghao/BeyondBoundingBox

Few-Shot目标检测

26. Accurate Few-Shot Object Detection With Support-Query Mutual Guidance and Hybrid Loss

• 作者单位: 复旦大学, 同济大学, 浙江大学

• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Zhang_Accurate_Few-Shot_Object_Detection_With_Support-Query_Mutual_Guidance_and_Hybrid_CVPR_2021_paper.html

• Code: None

27. Adaptive Image Transformer for One-Shot Object Detection

• 作者单位: 中央研究院, 台湾AI Labs
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Chen_Adaptive_Image_Transformer_for_One-Shot_Object_Detection_CVPR_2021_paper.html
• Code: None

28. Dense Relation Distillation with Context-aware Aggregation for Few-Shot Object Detection

• 作者单位: 北京大学, 北邮
• Paper: https://arxiv.org/abs/2103.17115
• Code: https://github.com/hzhupku/DCNet

29. Semantic Relation Reasoning for Shot-Stable Few-Shot Object Detection

• 作者单位: 卡内基梅隆大学(CMU)

• Paper: https://arxiv.org/abs/2103.01903

• Code: None

30. FSCE: Few-Shot Object Detection via Contrastive Proposal Encoding

• 作者单位: 南加利福尼亚大学, 旷视科技
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Sun_FSCE_Few-Shot_Object_Detection_via_Contrastive_Proposal_Encoding_CVPR_2021_paper.html
• Code: https://github.com/MegviiDetection/FSCE

31. Hallucination Improves Few-Shot Object Detection

• 作者单位: 伊利诺伊大学厄巴纳-香槟分校
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Zhang_Hallucination_Improves_Few-Shot_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/pppplin/HallucFsDet

32. Few-Shot Object Detection via Classification Refinement and Distractor Retreatment

• 作者单位: 新加坡国立大学, SIMTech
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Li_Few-Shot_Object_Detection_via_Classification_Refinement_and_Distractor_Retreatment_CVPR_2021_paper.html
• Code: None

33. Generalized Few-Shot Object Detection Without Forgetting

• 作者单位: 旷视科技
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Fan_Generalized_Few-Shot_Object_Detection_Without_Forgetting_CVPR_2021_paper.html
• Code: None

34. Transformation Invariant Few-Shot Object Detection

• 作者单位: 华为诺亚方舟实验室

• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Li_Transformation_Invariant_Few-Shot_Object_Detection_CVPR_2021_paper.html

• Code: None

35. UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation

• 作者单位: 不列颠哥伦比亚大学, Vector AI, CIFAR AI Chair
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Khandelwal_UniT_Unified_Knowledge_Transfer_for_Any-Shot_Object_Detection_and_Segmentation_CVPR_2021_paper.html
• Code: https://github.com/ubc-vision/UniT

36. Beyond Max-Margin: Class Margin Equilibrium for Few-Shot Object Detection

• 作者单位: 国科大, 厦门大学, 鹏城实验室
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Li_Beyond_Max-Margin_Class_Margin_Equilibrium_for_Few-Shot_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/Bohao-Lee/CME

半监督目标检测

37. Points As Queries: Weakly Semi-Supervised Object Detection by Points]

• 作者单位: 旷视科技, 复旦大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Chen_Points_As_Queries_Weakly_Semi-Supervised_Object_Detection_by_Points_CVPR_2021_paper.html
• Code: None

38. Data-Uncertainty Guided Multi-Phase Learning for Semi-Supervised Object Detection

• 作者单位: 清华大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Wang_Data-Uncertainty_Guided_Multi-Phase_Learning_for_Semi-Supervised_Object_Detection_CVPR_2021_paper.html
• Code: None

39. Positive-Unlabeled Data Purification in the Wild for Object Detection

• 作者单位: 华为诺亚方舟实验室, 悉尼大学, 北京大学

• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Guo_Positive-Unlabeled_Data_Purification_in_the_Wild_for_Object_Detection_CVPR_2021_paper.html

• Code: None

40. Interactive Self-Training With Mean Teachers for Semi-Supervised Object Detection

• 作者单位: 阿里巴巴, 香港理工大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Yang_Interactive_Self-Training_With_Mean_Teachers_for_Semi-Supervised_Object_Detection_CVPR_2021_paper.html
• Code: None

41. Instant-Teaching: An End-to-End Semi-Supervised Object Detection Framework

• 作者单位: 阿里巴巴
• Paper: https://arxiv.org/abs/2103.11402
• Code: None

42. Humble Teachers Teach Better Students for Semi-Supervised Object Detection

• 作者单位: 卡内基梅隆大学(CMU), 亚马逊
• Homepage: https://yihet.com/humble-teacher
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Tang_Humble_Teachers_Teach_Better_Students_for_Semi-Supervised_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/lryta/HumbleTeacher

43. Interpolation-Based Semi-Supervised Learning for Object Detection

• 作者单位: 首尔大学, 阿尔托大学等
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Jeong_Interpolation-Based_Semi-Supervised_Learning_for_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/soo89/ISD-SSD

域自适应目标检测

44. Domain-Specific Suppression for Adaptive Object Detection

• 作者单位: 中科院, 寒武纪, 国科大
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Wang_Domain-Specific_Suppression_for_Adaptive_Object_Detection_CVPR_2021_paper.html
• Code: None

45. MeGA-CDA: Memory Guided Attention for Category-Aware Unsupervised Domain Adaptive Object Detection

• 作者单位: 约翰斯·霍普金斯大学, 梅赛德斯—奔驰
• Paper: https://arxiv.org/abs/2103.04224
• Code: None

46. Unbiased Mean Teacher for Cross-Domain Object Detection

• 作者单位: 电子科技大学, ETH Zurich
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Deng_Unbiased_Mean_Teacher_for_Cross-Domain_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/kinredon/umt

47. I^3Net: Implicit Instance-Invariant Network for Adapting One-Stage Object Detectors

• 作者单位: 香港大学, 厦门大学, Deepwise AI Lab
• Paper: https://arxiv.org/abs/2103.13757
• Code: None

自监督目标检测

48. There Is More Than Meets the Eye: Self-Supervised Multi-Object Detection and Tracking With Sound by Distilling Multimodal Knowledge

• 作者单位: 弗莱堡大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Valverde_There_Is_More_Than_Meets_the_Eye_Self-Supervised_Multi-Object_Detection_CVPR_2021_paper.html
• Code: http://rl.uni-freiburg.de/research/multimodal-distill

49. Instance Localization for Self-supervised Detection Pretraining

• 作者单位: 香港中文大学, 微软亚洲研究院
• Paper: https://arxiv.org/abs/2102.08318
• Code: https://github.com/limbo0000/InstanceLoc

弱监督目标检测

50. Informative and Consistent Correspondence Mining for Cross-Domain Weakly Supervised Object Detection

• 作者单位: 北航, 鹏城实验室, 商汤科技
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Hou_Informative_and_Consistent_Correspondence_Mining_for_Cross-Domain_Weakly_Supervised_Object_CVPR_2021_paper.html
• Code: None

51. DAP: Detection-Aware Pre-training with Weak Supervision

• 作者单位: UIUC, 微软
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Zhong_DAP_Detection-Aware_Pre-Training_With_Weak_Supervision_CVPR_2021_paper.html
• Code: None

其他

52. Open-Vocabulary Object Detection Using Captions

• 作者单位：Snap, 哥伦比亚大学

• Paper(Oral): https://openaccess.thecvf.com/content/CVPR2021/html/Zareian_Open-Vocabulary_Object_Detection_Using_Captions_CVPR_2021_paper.html

• Code: https://github.com/alirezazareian/ovr-cnn

53. Depth From Camera Motion and Object Detection

• 作者单位: 密歇根大学, SIAI

• Paper: https://arxiv.org/abs/2103.01468

• Code: https://github.com/griffbr/ODMD

• Dataset: https://github.com/griffbr/ODMD

54. Unsupervised Object Detection With LIDAR Clues

• 作者单位: 商汤科技, 国科大, 中科大
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Tian_Unsupervised_Object_Detection_With_LIDAR_Clues_CVPR_2021_paper.html
• Code: None

55. GAIA: A Transfer Learning System of Object Detection That Fits Your Needs

• 作者单位: 国科大, 北理, 中科院, 商汤科技
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Bu_GAIA_A_Transfer_Learning_System_of_Object_Detection_That_Fits_CVPR_2021_paper.html
• Code: https://github.com/GAIA-vision/GAIA-det

56. General Instance Distillation for Object Detection

• 作者单位: 旷视科技, 北航
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Dai_General_Instance_Distillation_for_Object_Detection_CVPR_2021_paper.html
• Code: None

57. AQD: Towards Accurate Quantized Object Detection

• 作者单位: 蒙纳士大学, 阿德莱德大学, 华南理工大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Chen_AQD_Towards_Accurate_Quantized_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/aim-uofa/model-quantization

58. Scale-Aware Automatic Augmentation for Object Detection

• 作者单位: 香港中文大学, 字节跳动AI Lab, 思谋科技
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Chen_Scale-Aware_Automatic_Augmentation_for_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/Jia-Research-Lab/SA-AutoAug

59. Equalization Loss v2: A New Gradient Balance Approach for Long-Tailed Object Detection

• 作者单位: 同济大学, 商汤科技, 清华大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Tan_Equalization_Loss_v2_A_New_Gradient_Balance_Approach_for_Long-Tailed_CVPR_2021_paper.html
• Code: https://github.com/tztztztztz/eqlv2

60. Class-Aware Robust Adversarial Training for Object Detection

• 作者单位: 哥伦比亚大学, 中央研究院
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Chen_Class-Aware_Robust_Adversarial_Training_for_Object_Detection_CVPR_2021_paper.html
• Code: None

61. Improved Handling of Motion Blur in Online Object Detection

• 作者单位: 伦敦大学学院
• Homepage: http://visual.cs.ucl.ac.uk/pubs/handlingMotionBlur/
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Sayed_Improved_Handling_of_Motion_Blur_in_Online_Object_Detection_CVPR_2021_paper.html
• Code: None

62. Multiple Instance Active Learning for Object Detection

• 作者单位: 国科大, 华为诺亚
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Yuan_Multiple_Instance_Active_Learning_for_Object_Detection_CVPR_2021_paper.html
• Code: https://github.com/yuantn/MI-AOD

63. Neural Auto-Exposure for High-Dynamic Range Object Detection

• 作者单位: Algolux, 普林斯顿大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Onzon_Neural_Auto-Exposure_for_High-Dynamic_Range_Object_Detection_CVPR_2021_paper.html
• Code: None

64. Generalizable Pedestrian Detection: The Elephant in the Room

• 作者单位: IIAI, 阿尔托大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Hasan_Generalizable_Pedestrian_Detection_The_Elephant_in_the_Room_CVPR_2021_paper.html
• Code: https://github.com/hasanirtiza/Pedestron

65. Neural Auto-Exposure for High-Dynamic Range Object Detection

• 作者单位: Algolux, 普林斯顿大学
• Paper: https://openaccess.thecvf.com/content/CVPR2021/html/Onzon_Neural_Auto-Exposure_for_High-Dynamic_Range_Object_Detection_CVPR_2021_paper.html
• Code: None

CVer-目标检测交流群

建了CVer-目标检测交流群！想要进目标检测学习交流群的同学，可以直接加微信号：CVer9999。加的时候备注一下：目标检测+学校/公司+昵称，即可。然后就可以拉你进群了。

强烈推荐大家关注CVer知乎账号和CVer微信公众号，可以快速了解到最新优质的CV论文。