视频目标检测--Flow-Guided Feature Aggregation for Video Object Detection

Flow-Guided Feature Aggregation for Video Object Detection
https://arxiv.org/abs/1703.10025
Our framework is principled, and on par with the best engineered systems winning the ImageNet VID challenges 2016

The code would be released

本文主要利用视频中前后帧的特征信息来提高当前帧的目标检测精度。
we propose to improve the per-frame feature learning by temporal aggregation

为什么需要前后帧信息了，因为视频中有时候每一帧的目标信息不是适合于检测

Note that the features of the same object instance are usually not spatially aligned across frames due to video motion.
前后帧的特征信息怎么对应起来了？两个模块：1）基于运动指导的 spatial warping 2）特征融合模块
Two modules are necessary for such feature propagation and enhancement:
1) motion-guided spatial warping. It estimates the motion between frames and warps the feature maps accordingly.
2) feature aggregation module. It figures out how to properly fuse the features from multiple frames.

3.2. Model Design
Flow-guided warping
首先用一个 flow network F 估计光流场，然后基于光流场把当前帧的特征映射到相邻帧去，得到相邻帧的特征图 bilinear warping

Feature aggregation
有了当前帧的特征图及其前后相邻若干帧的特征图，我们采用 Adaptive weight 来融合这些特征图得到最终的特征图，将该特征图输入检测子网络中

总的计算流程

不同模块组合的 Accuracy and runtime

加入 flow-guided 时间增加的比较多

小目标的检测还是有待加强

视频目标检测--Flow-Guided Feature Aggregation for Video Object Detection相关推荐

【视频目标检测论文阅读笔记】Optimizing Video Object Detection via a Scale-Time Lattice
1.1 论文信息标题 Optimizing Video Object Detection via a Scale-Time Lattice 会议 CVPR 2018 原文链接 Optimizing ...
ICCV 2017 《Flow-Guided Feature Aggregation for Video Object Detection》论文笔记
本学弱喜欢在本子上记笔记,但字迹又丑. 望看不懂我的字的大佬不要喷我,看得懂的大佬批评指正.
目标检测--Beyond Skip Connections: Top-Down Modulation for Object Detection
Beyond Skip Connections: Top-Down Modulation for Object Detection CVPR2017 under review code :(comin ...
目标检测 | RetinaNet：Focal Loss for Dense Object Detection
论文分析了 one-stage 网络训练存在的类别不平衡问题,提出能根据 loss 大小自动调节权重的 focal loss,使得模型的训练更专注于困难样本.同时,基于 FPN 设计了 RetinaN ...
【目标检测】61、Dynamic Head Unifying Object Detection Heads with Attentions
文章目录一.背景二.方法 2.1 scale-aware attention 2.2 spatial-aware attention 2.3 task-aware attention 2.4 总体 ...
目标检测（二）--Hough Forests for Object Detection
原文: http://www.cnblogs.com/walccott/p/4956860.html 09年cvpr上提出霍夫森林的文章--Class-Specific Hough Forests ...
【3D目标检测】WEAKM3D: TOWARDS WEAKLY SUPERVISEDMONOCULAR 3D OBJECT DETECTION
目录概述细节背景算法流程最终损失几何目标点云对齐损失射线追踪损失基于点密度的点级损失平衡边界框参数预测的解耦概述 [2022][WEAKM3D] 研究的问题: 核心问题:如何以2D ...
Stitcher-小目标检测算法 | Stitcher: Feedback-driven Data Provider for Object Detection
刚出来的小目标检测论文论文地址:https://arxiv.org/pdf/2004.12432.pdf Abstract: 目标检测器通常会根据尺寸不同具有不同性能表现,其中小物体的性能最不令人满 ...
小目标检测论文笔记：Learning Calibrated-Guidance for Object Detection in Aerial Images
论文 Learning Calibrated-Guidance for Object Detection in Aerial Images paper:https://arxiv.org/abs/21 ...

视频目标检测--Flow-Guided Feature Aggregation for Video Object Detection

视频目标检测--Flow-Guided Feature Aggregation for Video Object Detection相关推荐

最新文章

热门文章