ECCV 2022|华为诺亚提出CLIFF：将全帧位置信息带入人体姿势和形状估！论文/代码速递2022.10.18！

整理：AI算法与图像处理

CVPR2022论文和代码整理：https://github.com/DWCTOD/CVPR2022-Papers-with-Code-Demo

ECCV2022论文和代码整理：https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo

最新成果demo展示：

标题：CLIFF: Carrying Location Information in Full Frames into Human Pose and Shape Estimation

代码：https://github.com/huawei-noah/noah-research/tree/master/CLIFF

论文：https://arxiv.org/abs/2208.00571

自顶向下的方法在3D人体姿势和形状估计领域占据主导地位，因为它们与人体检测分离，允许研究人员专注于核心问题。然而，裁剪是它们的第一步，从一开始就丢弃了位置信息，这使得它们无法在原始相机坐标系中准确预测全局旋转。为了解决这个问题，我们建议在这个任务中携带全帧位置信息（CLIFF）。具体来说，我们通过将裁剪的图像特征与其边界框信息连接起来，向CLIFF提供更全面的特征。我们在更宽的全帧视野下计算2D重投影损失，采用与在图像中投影的人相似的投影过程。CLIFF由全球位置感知信息提供并监督，它直接预测全球旋转以及更精确的关节姿势。此外，我们提出了一种基于CLIFF的伪地面真值注释器，它为野外二维数据集提供高质量的三维注释，并为基于回归的方法提供关键的全面监督。对流行基准测试的大量实验表明，CLIFF的表现明显优于现有技术，并在AGORA排行榜上排名第一（SMPL算法跟踪）。

最新论文整理

ECCV2022

Updated on : 18 Oct 2022

total number : 14

CramNet: Camera-Radar Fusion with Ray-Constrained Cross-Attention for Robust 3D Object Detection

• 论文/Paper: http://arxiv.org/pdf/2210.09267
• 代码/Code: None

Improving Contrastive Learning on Visually Homogeneous Mars Rover Images

• 论文/Paper: http://arxiv.org/pdf/2210.09234
• 代码/Code: None

ArtFacePoints: High-resolution Facial Landmark Detection in Paintings and Prints

• 论文/Paper: http://arxiv.org/pdf/2210.09204
• 代码/Code: https://github.com/asindel/artfacepoints

Distilling Object Detectors With Global Knowledge

• 论文/Paper: http://arxiv.org/pdf/2210.09022
• 代码/Code: https://github.com/hikvision-research/davar-lab-ml

AIM 2022 Challenge on Instagram Filter Removal: Methods and Results

• 论文/Paper: http://arxiv.org/pdf/2210.08997
• 代码/Code: None

Rethinking Trajectory Prediction via "Team Game"

• 论文/Paper: http://arxiv.org/pdf/2210.08793
• 代码/Code: None

Temporal and Contextual Transformer for Multi-Camera Editing of TV Shows

• 论文/Paper: http://arxiv.org/pdf/2210.08737
• 代码/Code: None

Runner-Up Solution to Google Universal Image Embedding Competition 2022

• 论文/Paper: http://arxiv.org/pdf/2210.08735
• 代码/Code: None

Selective Query-guided Debiasing Network for Video Corpus Moment Retrieval

• 论文/Paper: http://arxiv.org/pdf/2210.08714
• 代码/Code: None

1st Place Solution in Google Universal Images Embedding

• 论文/Paper: http://arxiv.org/pdf/2210.08473
• 代码/Code: None

Video in 10 Bits: Few-Bit VideoQA for Efficiency and Privacy

• 论文/Paper: http://arxiv.org/pdf/2210.08391
• 代码/Code: None

Improving the Intra-class Long-tail in 3D Detection via Rare Example Mining

• 论文/Paper: http://arxiv.org/pdf/2210.08375
• 代码/Code: None

Geometric Representation Learning for Document Image Rectification

• 论文/Paper: http://arxiv.org/pdf/2210.08161
• 代码/Code: https://github.com/fh2019ustc/docgeonet

Motion Inspired Unsupervised Perception and Prediction in Autonomous Driving

• 论文/Paper: http://arxiv.org/pdf/2210.08061
• 代码/Code: None

CVPR2022

NeurIPS

Updated on : 18 Oct 2022

total number : 12

S^3-NeRF: Neural Reflectance Field from Shading and Shadow under a Single Viewpoint

• 论文/Paper: http://arxiv.org/pdf/2210.08936
• 代码/Code: None

Contrastive Language-Image Pre-Training with Knowledge Graphs

• 论文/Paper: http://arxiv.org/pdf/2210.08901
• 代码/Code: None

HyperDomainNet: Universal Domain Adaptation for Generative Adversarial Networks

• 论文/Paper: http://arxiv.org/pdf/2210.08884
• 代码/Code: https://github.com/macderru/hyperdomainnet

Scaling & Shifting Your Features: A New Baseline for Efficient Model Tuning

• 论文/Paper: http://arxiv.org/pdf/2210.08823
• 代码/Code: https://github.com/dongzelian/ssf

Signal Processing for Implicit Neural Representations

• 论文/Paper: http://arxiv.org/pdf/2210.08772
• 代码/Code: None

Forecasting Human Trajectory from Scene History

• 论文/Paper: http://arxiv.org/pdf/2210.08732
• 代码/Code: None

LAION-5B: An open large-scale dataset for training next generation image-text models

• 论文/Paper: http://arxiv.org/pdf/2210.08402
• 代码/Code: None

mRI: Multi-modal 3D Human Pose Estimation Dataset using mmWave, RGB-D, and Inertial Sensors

• 论文/Paper: http://arxiv.org/pdf/2210.08394
• 代码/Code: None

Self-Supervised Learning Through Efference Copies

• 论文/Paper: http://arxiv.org/pdf/2210.09224
• 代码/Code: None

Marksman Backdoor: Backdoor Attacks with Arbitrary Target Class

• 论文/Paper: http://arxiv.org/pdf/2210.09194
• 代码/Code: None

Navigating Memory Construction by Global Pseudo-Task Simulation for Continual Learning

• 论文/Paper: http://arxiv.org/pdf/2210.08442
• 代码/Code: https://github.com/liuyejia/gps_cl

Neural Attentive Circuits

• 论文/Paper: http://arxiv.org/pdf/2210.08031
• 代码/Code: None