精品基于深度神经网络的目标检测.pptVIP

第 1 页 | 共 25 页 基于深度神经网络的目标检测 检测 ≈ 定位 + 分类 第 2 页 | 共 25 页 第 3 页 | 共 25 页 传统目标检测 基于候选区域的 目标检测 回归方法的深度 学习目标检测 faster-RCNN fast-RCNN SPPNET RCNN YOLO 第 4 页 | 共 25 页 区域选择 特征提取 分类器分类 滑动窗口策略 手工设计特征 SVM 、 adaboost 等 两个问题 传统目标检测 第 5 页 | 共 25 页 候选区域 selective search IOU NMS 非极大值抑制 PASCAL VOC 上的检测率 从 35.1% 提升到 53.7% 基于候选区域的目标检测 第 6 页 | 共 25 页 Step1. Input an image Step2. Use selective search to obtain ~2k proposals Step3. Warp each proposal and apply CNN to extract its features Step4. Adopt class-specified SVM to score each proposal Step5. Rank the proposals and use NMS to get the bboxes . Step6. Use class-specified regressors to refine the bboxes positions. TOO SLOWWWW !!! － SPPNET RCNN 三个问题 （分阶段训练、 空间浪费、慢 47s ） SPP-Net : Motivation Cropping may loss some information about the object Warpping may change the object s appearance 第 7 页 | 共 25 页 ? FC layer need a fixed-length input while conv layer can be adapted to arbitrary input size. ? Thus we need a bridge between the conv and FC layer. ? Here comes the SPP layer. 第 8 页 | 共 25 页 SPP-Net: Training for Detection(1) 第 9 页 | 共 25 页 Conv5 feature map Conv5 feature map Conv5 feature map Image Pyramid FeatMap Pyramids conv Step1. Generate a image pyramid and exact the conv FeatMap of the whole image SPP-Net: Training for Detection(2) ? Step 2, For each proposal, walking the image pyramid and find a project version that has a number of pixels closest to 224x224. (For scaling invariance in training.) ? Step 3, find the corresponding FeatMap in Conv5 and use SPP layer to pool it to a fix size. ? Step 4, While getting all the proposals feature, fine-tune the FC layer only . ? Step 5, Train the class- specified SVM 第 10 页 | 共 25 页 SPP-Net: Testing for Detection Almost the same as R-CNN, except Step3. 第 11 页 | 共 25 页 第 12 页 | 共 25 页 ? Speed: 64x faster than R-CNN using one scale, and 24x faster using five-scale paramid. ? mAP: +1.2 mAP vs R-CNN 2. 训练花费过大的硬盘开销和时间 1. 训练分多阶段，并不是端到端的训练过程 FC