ECCV&CVPR論文速遞2022.7.5！最新成果demo展示
最新成果demo展示：
ECCV2022 | FlowFormer: 基于transformer架構(gòu)的光流
標(biāo)題：FlowFormer: A Transformer Architecture for Optical Flow 論文：https://arxiv.org/abs/2203.16194
主頁：https://drinkingcoder.github.io/publication/flowformer/
摘要：本文介紹了光流transformer（FlowFormer），一種基于transformer的神經(jīng)網(wǎng)絡(luò)架構(gòu)，用于學(xué)習(xí)光流。FlowFormer 對從圖像對構(gòu)建的 4D 成本量進行標(biāo)記，將cost token 編碼到具有新穎潛在空間中的交替組transformer(AGT) 層的成本存儲器中，并通過具有動態(tài)位置成本查詢的循環(huán)變換器解碼器對成本存儲器進行解碼 . 在 Sintel 基準(zhǔn)測試中，F(xiàn)lowFormer 實現(xiàn)了 1.178 的平均端點誤差 (AEPE)，與公布的最佳結(jié)果 (1.388) 相比，誤差減少了 15.1%。此外，F(xiàn)lowFormer 還實現(xiàn)了強大的泛化性能。在沒有接受 Sintel 訓(xùn)練的情況下，F(xiàn)lowFormer 在 Sintel 訓(xùn)練集干凈通行證上達到 1.00 AEPE，比公布的最佳結(jié)果 (1.29) 高出 22.4%。
最新論文整理：
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
論文/Paper: http://arxiv.org/pdf/2207.01573
代碼/Code: None
Open-world Semantic Segmentation for LIDAR Point Clouds
論文/Paper: http://arxiv.org/pdf/2207.01452
代碼/Code: https://github.com/jun-cen/open_world_3d_semantic_segmentation
GraphVid: It Only Takes a Few Nodes to Understand a Video
論文/Paper: http://arxiv.org/pdf/2207.01375
代碼/Code: None
Target-absent Human Attention
論文/Paper: http://arxiv.org/pdf/2207.01166
代碼/Code: None
Lottery Ticket Hypothesis for Spiking Neural Networks
論文/Paper: http://arxiv.org/pdf/2207.01382
代碼/Code: None
CVPR2022 Updated on : 5 Jul 2022
total number : 6
Task Discrepancy Maximization for Fine-grained Few-Shot Classification
論文/Paper: http://arxiv.org/pdf/2207.01376
代碼/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
論文/Paper: http://arxiv.org/pdf/2207.01334
代碼/Code: https://github.com/showlab/egovlp
Aug-NeRF: Training Stronger Neural Radiance Fields with Triple-Level Physically-Grounded Augmentations
論文/Paper: http://arxiv.org/pdf/2207.01164
代碼/Code: https://github.com/vita-group/aug-nerf
PhotoScene: Photorealistic Material and Lighting Transfer for Indoor Scenes
論文/Paper: http://arxiv.org/pdf/2207.00757
代碼/Code: https://github.com/vilab-ucsd/photoscene
Golfer: Trajectory Prediction with Masked Goal Conditioning MnM Network
論文/Paper: http://arxiv.org/pdf/2207.00738
代碼/Code: None
DRESS: Dynamic REal-time Sparse Subnets
論文/Paper: http://arxiv.org/pdf/2207.00670
代碼/Code: None