最新成果demo展示：

ECCV2022 | FlowFormer: 基于transformer架构的光流

标题：FlowFormer: A Transformer Architecture for Optical Flow 论文：https://arxiv.org/abs/2203.16194

主页：https://drinkingcoder.github.io/publication/flowformer/

摘要：本文介绍了光流transformer（FlowFormer），一种基于transformer的神经网络架构，用于学习光流。FlowFormer 对从图像对构建的 4D 成本量进行标记，将cost token 编码到具有新颖潜在空间中的交替组transformer(AGT) 层的成本存储器中，并通过具有动态位置成本查询的循环变换器解码器对成本存储器进行解码 . 在 Sintel 基准测试中，FlowFormer 实现了 1.178 的平均端点误差 (AEPE)，与公布的最佳结果 (1.388) 相比，误差减少了 15.1%。此外，FlowFormer 还实现了强大的泛化性能。在没有接受 Sintel 训练的情况下，FlowFormer 在 Sintel 训练集干净通行证上达到 1.00 AEPE，比公布的最佳结果 (1.29) 高出 22.4%。

最新论文整理：

ECCV 2022 Updated on :  5 Jul 2022

total number : 5

Embedding contrastive unsupervised features to cluster in- and out-of-distribution noise in corrupted image datasets

• 论文/Paper: http://arxiv.org/pdf/2207.01573

• 代码/Code: None

Open-world Semantic Segmentation for LIDAR Point Clouds

• 论文/Paper: http://arxiv.org/pdf/2207.01452

• 代码/Code: https://github.com/jun-cen/open_world_3d_semantic_segmentation

GraphVid: It Only Takes a Few Nodes to Understand a Video

• 论文/Paper: http://arxiv.org/pdf/2207.01375

• 代码/Code: None

Target-absent Human Attention

• 论文/Paper: http://arxiv.org/pdf/2207.01166

• 代码/Code: None

Lottery Ticket Hypothesis for Spiking Neural Networks

• 论文/Paper: http://arxiv.org/pdf/2207.01382

• 代码/Code: None

CVPR2022 Updated on :  5 Jul 2022

total number : 6

Task Discrepancy Maximization for Fine-grained Few-Shot Classification

• 论文/Paper: http://arxiv.org/pdf/2207.01376

• 代码/Code: https://github.com/leesb7426/cvpr2022-task-discrepancy-maximization-for-fine-grained-few-shot-classification

Egocentric Video-Language Pretraining @ EPIC-KITCHENS-100 Multi-Instance Retrieval Challenge 2022

• 论文/Paper: http://arxiv.org/pdf/2207.01334

• 代码/Code: https://github.com/showlab/egovlp

Aug-NeRF: Training Stronger Neural Radiance Fields with Triple-Level Physically-Grounded Augmentations

• 论文/Paper: http://arxiv.org/pdf/2207.01164

• 代码/Code: https://github.com/vita-group/aug-nerf

PhotoScene: Photorealistic Material and Lighting Transfer for Indoor Scenes

• 论文/Paper: http://arxiv.org/pdf/2207.00757

• 代码/Code: https://github.com/vilab-ucsd/photoscene

Golfer: Trajectory Prediction with Masked Goal Conditioning MnM Network

• 论文/Paper: http://arxiv.org/pdf/2207.00738

• 代码/Code: None

DRESS: Dynamic REal-time Sparse Subnets

• 论文/Paper: http://arxiv.org/pdf/2207.00670

• 代码/Code: None