ECCV2022 | 谷歌 AR 团队最新成果!论文速递2022.10.6!

2022-12-11 12:47:16 浏览数 (1)

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:

Learned Monocular Depth Priors in Visual-Inertial Initialization

论文:https://arxiv.org/abs/2204.09171

代码:

摘要:

视觉惯性里程表(VIO)是当今学术界和工业界大多数AR/VR和自主机器人系统的姿态估计支柱。然而,这些系统对传感器偏差、重力方向和公制刻度等关键参数的初始化高度敏感。在很少满足高视差或可变加速度假设的实际场景中(例如,悬停空中机器人、智能手机AR用户不使用手机手势),经典的视觉惯性初始化公式通常会出现病态和/或无法有效收敛。在本文中,我们专门针对这些对野外使用至关重要的低激励场景进行视觉惯性初始化。我们建议通过将一种新的基于学习的测量作为更高级别的输入,来绕过传统视觉惯性结构运动(SfM)初始化的局限性。我们利用学习的单目深度图像(单目深度)来限制特征的相对深度,并通过联合优化其比例和偏移将单目深度升级为公制比例。我们的实验表明,与视觉惯性初始化的经典公式相比,问题条件处理有了显著改善,并且与公共基准的最新水平相比,尤其是在低激励情况下,精度和鲁棒性有了显著提高。我们进一步将此改进扩展到现有里程表系统中的实现,以说明改进的初始化方法对结果跟踪轨迹的影响。


最新论文整理

ECCV2022

Updated on : 6 Oct 2022
total number : 3

Granularity-aware Adaptation for Image Retrieval over Multiple Tasks

  • 论文/Paper: http://arxiv.org/pdf/2210.02254
  • 代码/Code: None

Dfferentiable Raycasting for Self-supervised Occupancy Forecasting

  • 论文/Paper: http://arxiv.org/pdf/2210.01917
  • 代码/Code: https://github.com/tarashakhurana/emergent-occ-forecasting

A Perceptual Quality Metric for Video Frame Interpolation

  • 论文/Paper: http://arxiv.org/pdf/2210.01879
  • 代码/Code: https://github.com/hqqxyy/vfips

CVPR2022

NeurIPS

Updated on : 6 Oct 2022
total number : 6

Promising or Elusive? Unsupervised Object Segmentation from Real-world Single Images

  • 论文/Paper: http://arxiv.org/pdf/2210.02324
  • 代码/Code: https://github.com/vlar-group/unsupobjseg

Weak-shot Semantic Segmentation via Dual Similarity Transfer

  • 论文/Paper: http://arxiv.org/pdf/2210.02270
  • 代码/Code: https://github.com/bcmi/SimFormer-Weak-Shot-Semantic-Segmentation.

Relational Proxies: Emergent Relationships as Fine-Grained Discriminators

  • 论文/Paper: http://arxiv.org/pdf/2210.02149
  • 代码/Code: https://github.com/abhrac/relational-proxies

Natural Color Fool: Towards Boosting Black-box Unrestricted Attacks

  • 论文/Paper: http://arxiv.org/pdf/2210.02041
  • 代码/Code: https://github.com/ylhz/Natural-Color-Fool.

GMMSeg: Gaussian Mixture based Generative Semantic Segmentation Models

  • 论文/Paper: http://arxiv.org/pdf/2210.02025
  • 代码/Code: https://github.com/leonnnop/GMMSeg

On the Learning Mechanisms in Physical Reasoning

  • 论文/Paper: http://arxiv.org/pdf/2210.02075
  • 代码/Code: None

0 人点赞