Faster R-CNN: Towards real-time object detection with region proposal networks
说明： introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals。An RPN is a fully-convolutional network that simultaneously predicts object bounds and objectness scores at each position.
Introduction
1，region proposals are the computational bottleneck in state-of-the-art detection systems。
Region Proposal Networks
1，A Region Proposal Network (RPN) takes an image (of any size) as input and outputs a set of rectangular object proposals, each with an objectness score。
2，This network is fully connected to an n × n spatial window of the input conv feature map. Each sliding window is mapped to a lower-dimensional vector (256-d for ZF and 512-d for VGG). This vector is fed into two sibling fully-connected layers—a box-regression layer (reg) and a box-classification layer (cls)。
3，predict k region proposals, so the reg layer，has 4k outputs encoding the coordinates of k boxes。
A Loss Function for Learning Region Proposals
assign a positive label to two kinds of anchors: (i) the anchor/anchors with the highest Intersectionover-Union (IoU) overlap with a ground-truth box, or (ii) an anchor that has an IoU overlap higher than 0.7 with any ground-truth box.
Optimization
1，The RPN, which is naturally implemented as a fully-convolutional network, can be trained end-to-end by back-propagation and stochastic gradient descent (SGD) 。
2，we randomly sample 256 anchors in an image to compute the loss function of a mini-batch, where the sampled positive and negative anchors have a ratio of up to 1:1。
3，all new layers的权值初始化：高斯分布(μ=0,σ=0.01)，all other layers（比如共享卷积层）用ImageNet来权值初始化。用ZF net来进行进行微调。学习率：0.001(60k)->0.0001(20k)。动量：0.9。weight decay: 0.0005。
Sharing Convolutional Features for Region Proposal and Object Detection
1，sharing convolutional layers between the two networks, rather than learning two separate networks
2， 4-step training algorithm to learn shared features via alternating optimization
Implementation Details
1，Multi-scale与speed-accuracy之间的trade-off
2，To reduce redundancy, we adopt non-maximum suppression (NMS) on the proposal regions based on their cls scores.