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：
主页：https://sgvr.kaist.ac.kr/publication/flow-supervisor/ 代码：https://github.com/iwbn/flow-supervisor
光流CNN的训练管道由合成数据集的预训练阶段和目标数据集的微调阶段组成。然而，从目标视频中获取ground truth 流需要付出巨大的努力。本文提出了一种实用的微调方法，以使预处理模型适应没有ground truth 流的目标数据集，这种方法尚未得到广泛的探索。具体来说，我们提出了一个用于自监督的流监督，它由参数分离和学生输出连接组成。这种设计的目的是稳定收敛，并比在微调任务中不稳定的传统自监督方法具有更好的精度。实验结果表明，与不同的自监督方法相比，该方法对于半监督学习是有效的。此外，通过利用额外的未标记数据集，我们在Sintel和KITTI基准上对最先进的光流模型进行了有意义的改进
最新论文整理
ECCV2022
Updated on : 12 Oct 2022
total number : 5
Oflib: Facilitating Operations with and on Optical Flow Fields in Python
- 论文/Paper: http://arxiv.org/pdf/2210.05635
- 代码/Code: None
Global Spectral Filter Memory Network for Video Object Segmentation
- 论文/Paper: http://arxiv.org/pdf/2210.05567
- 代码/Code: https://github.com/workforai/gsfm
Map-free Visual Relocalization: Metric Pose Relative to a Single Image
- 论文/Paper: http://arxiv.org/pdf/2210.05494
- 代码/Code: https://github.com/nianticlabs/map-free-reloc
LidarNAS: Unifying and Searching Neural Architectures for 3D Point Clouds
- 论文/Paper: http://arxiv.org/pdf/2210.05018
- 代码/Code: None
Graph2Vid: Flow graph to Video Grounding forWeakly-supervised Multi-Step Localization
- 论文/Paper: http://arxiv.org/pdf/2210.04996
- 代码/Code: None
CVPR2022
NeurIPS
Updated on : 12 Oct 2022
total number : 9
Point Transformer V2: Grouped Vector Attention and Partition-based Pooling
- 论文/Paper: http://arxiv.org/pdf/2210.05666
- 代码/Code: https://github.com/gofinge/pointtransformerv2
The Unreasonable Effectiveness of Fully-Connected Layers for Low-Data Regimes
- 论文/Paper: http://arxiv.org/pdf/2210.05657
- 代码/Code: None
Neural Shape Deformation Priors
- 论文/Paper: http://arxiv.org/pdf/2210.05616
- 代码/Code: None
Prototypical VoteNet for Few-Shot 3D Point Cloud Object Detection
- 论文/Paper: http://arxiv.org/pdf/2210.05593
- 代码/Code: https://github.com/cvmi-lab/fs3d
FreGAN: Exploiting Frequency Components for Training GANs under Limited Data
- 论文/Paper: http://arxiv.org/pdf/2210.05461
- 代码/Code: https://github.com/kobeshegu/fregan_neurips2022
Deep Fourier Up-Sampling
- 论文/Paper: http://arxiv.org/pdf/2210.05171
- 代码/Code: None
Learning with an Evolving Class Ontology
- 论文/Paper: http://arxiv.org/pdf/2210.04993
- 代码/Code: None
Make Sharpness-Aware Minimization Stronger: A Sparsified Perturbation Approach
- 论文/Paper: http://arxiv.org/pdf/2210.05177
- 代码/Code: https://github.com/mi-peng/sparse-sharpness-aware-minimization
VER: Scaling On-Policy RL Leads to the Emergence of Navigation in Embodied Rearrangement
- 论文/Paper: http://arxiv.org/pdf/2210.05064
- 代码/Code: None
