最新成果demo展示：

ECCV2022 | VNext：下一代视频实例识别框架

论文：https://arxiv.org/abs/2207.10661

代码：https://github.com/wjf5203/VNext 

ECCV2022 汇总：https://github.com/DWCTOD/ECCV2022-Papers-with-Code-Demo/

摘要：

近年来，视频实例分割（VIS）在很大程度上是由离线模型推动的，而在线模型通常比同时代的离线模型差 10 多个 AP，这是一个巨大的缺点。通过剖析当前的在线模型和离线模型，我们证明了性能差距的主要原因是容易出错的关联，并提出了 IDOL，它在三个基准上优于所有在线和离线方法。IDOL在第四届大规模视频对象分割挑战赛（CVPR2022）的视频实例分割赛道上获得第一名。

最新论文整理

   ECCV2022

Updated on : 2 Aug 2022

total number : 20

Video Question Answering with Iterative Video-Text Co-Tokenization

• 论文/Paper: http://arxiv.org/pdf/2208.00934

• 代码/Code: None

S$^2$Contact: Graph-based Network for 3D Hand-Object Contact Estimation with Semi-Supervised Learning

• 论文/Paper: http://arxiv.org/pdf/2208.00874

• 代码/Code: None

Skeleton-free Pose Transfer for Stylized 3D Characters

• 论文/Paper: http://arxiv.org/pdf/2208.00790

• 代码/Code: None

Deep 360$^\circ$ Optical Flow Estimation Based on Multi-Projection Fusion

• 论文/Paper: http://arxiv.org/pdf/2208.00776

• 代码/Code: None

Cross Attention Based Style Distribution for Controllable Person Image Synthesis

• 论文/Paper: http://arxiv.org/pdf/2208.00712

• 代码/Code: None

Improving Fine-Grained Visual Recognition in Low Data Regimes via Self-Boosting Attention Mechanism

• 论文/Paper: http://arxiv.org/pdf/2208.00617

• 代码/Code: https://github.com/GANPerf/SAM

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

• 论文/Paper: http://arxiv.org/pdf/2208.00571

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

SdAE: Self-distillated Masked Autoencoder

• 论文/Paper: http://arxiv.org/pdf/2208.00449

• 代码/Code: https://github.com/AbrahamYabo/SdAE.

Out-of-Distribution Detection with Semantic Mismatch under Masking

• 论文/Paper: http://arxiv.org/pdf/2208.00446

• 代码/Code: https://github.com/cure-lab/MOODCat

Toward Understanding WordArt: Corner-Guided Transformer for Scene Text Recognition

• 论文/Paper: http://arxiv.org/pdf/2208.00438

• 代码/Code: https://github.com/xdxie/WordArt

Skeleton-Parted Graph Scattering Networks for 3D Human Motion Prediction

• 论文/Paper: http://arxiv.org/pdf/2208.00368

• 代码/Code: None

Doubly Deformable Aggregation of Covariance Matrices for Few-shot Segmentation

• 论文/Paper: http://arxiv.org/pdf/2208.00306

• 代码/Code: None

Revisiting the Critical Factors of Augmentation-Invariant Representation Learning

• 论文/Paper: http://arxiv.org/pdf/2208.00275

• 代码/Code: None

RBP-Pose: Residual Bounding Box Projection for Category-Level Pose Estimation

• 论文/Paper: http://arxiv.org/pdf/2208.00237

• 代码/Code: None

Few-shot Single-view 3D Reconstruction with Memory Prior Contrastive Network

• 论文/Paper: http://arxiv.org/pdf/2208.00183

• 代码/Code: None

Few-Shot Class-Incremental Learning from an Open-Set Perspective

• 论文/Paper: http://arxiv.org/pdf/2208.00147

• 代码/Code: None

DAS: Densely-Anchored Sampling for Deep Metric Learning

• 论文/Paper: http://arxiv.org/pdf/2208.00119

• 代码/Code: https://github.com/lizhaoliu-Lec/DAS

Neural Correspondence Field for Object Pose Estimation

• 论文/Paper: http://arxiv.org/pdf/2208.00113

• 代码/Code: None

Explicit Occlusion Reasoning for Multi-person 3D Human Pose Estimation

• 论文/Paper: http://arxiv.org/pdf/2208.00090

• 代码/Code: None

Fast Two-step Blind Optical Aberration Correction

• 论文/Paper: http://arxiv.org/pdf/2208.00950

• 代码/Code: None

   CVPR2022

Updated on : 2 Aug 2022

total number : 2

Large-Scale Product Retrieval with Weakly Supervised Representation Learning

• 论文/Paper: http://arxiv.org/pdf/2208.00955

• 代码/Code: None

Generating Complex 4D Expression Transitions by Learning Face Landmark Trajectories

• 论文/Paper: http://arxiv.org/pdf/2208.00050

• 代码/Code: None