
极市导读
本文重点关注旷视在ECCV2022被接受的20篇论文,并整理出每篇论文的模型设计和亮点解读。论文内容涵盖了目标检测、3D重建、图像复原等多个方向。 >>加入极市CV技术交流群,走在计算机视觉的最前沿
01 Oral:Synergistic Self-Supervised and Quantization Learning
自监督学习与量化协同互助

02 Real-Time Intermediate Flow Estimation for Video Frame Interpolation
视频插帧中的实时中间流估计

03 KD-MVS: Knowledge Distillation Based Self-supervised Learning for Multi-view Stereo
KD-MVS: 基于知识蒸馏的自监督多视图立体网络训练方法

04 Sobolev Training for Implicit Neural Representations with Approximated Image Derivatives
基于近似图像梯度的隐式神经表达的索伯列夫训练

05 W2N: Switching From Weak Supervision to Noisy Supervision for Object Detection
W2N: 一种将弱监督信号转化为噪声监督信号的目标检测方法

06 Improving Image Restoration by Revisiting Global Information Aggregation
通过重新审视全局信息聚合来改善图像修复效果

07 Simple Baselines for Image Restoration
图像复原的简单基线

08 Revisiting the Critical Factors of Augmentation-Invariant Representation Learning
重新探索基于增广不变特征学习框架中的关键因素

09 PETR: Position Embedding Transformation for Multi-View 3D Object Detection
基于3D位置编码的多视角3D目标检测

10 MOTR: End-to-End Multiple-Object Tracking with TRansformer
基于Transformer的端到端多目标追踪框架

11 Oral:Tracking Objects as Pixel-wise Distributions
像素级别的多目标跟踪方案

12 Explaining Deepfake Detection by Analysing Image Matching
基于图像匹配解释深伪检测模型
① 深伪检测模型将伪造图像中同源图像和目标图像特征无关的图像区域判定为 Artifact 区域; ② 深伪检测模型在训练中同时借助二分类标签及数据中的图像匹配关系,隐式建模 Artifact 特征; ③ 深伪检测模型隐式建模的 Artifact 特征容易受到图像质量影响,导致模型压缩鲁棒性不足。

13 Motion Sensitive Contrastive Learning for Self-supervised Video Representation
用于自监督视频表示的运动敏感对比学习

14 Efficient One Pass Self-distillation with Zipf's Label Smoothing
基于Zipf分布实现标签平滑的高效一阶段自蒸馏方法

15 MegBA: A GPU-based Distributed Library for Large-Scale Bundle Adjustment
MegBA: 用于求解大规模Bundle Adjustment的GPU实现的分布式库

16 Oral:RealFlow: EM-based Realistic Optical Flow Datasets Generation from Videos
真实流: 基于EM算法从视频中生成真实光流数据集

17 Ghost-free High Dynamic Range Imaging with Context-aware Transformer
基于上下文感知Transformer的无鬼影高动态范围成像方法

18 D2C-SR: A Divergence to Convergence Approach for Real-World Image Super-Resolution
D2C-SR:一种针对真实世界超分辨率的发散到收敛方法

19 Discriminability-Transferability Trade-Off: An Information-Theoretic Perspective
以信息论的视角分析可判别性与可迁移性的权衡

20 Dense Teacher: Dense Pseudo-Labels for Semi-supervised Object Detection
密集学习:用于半监督学习的密集伪标签

公众号后台回复“ECCV2022”获取论文分类资源下载~

“
点击阅读原文进入CV社区
收获更多技术干货