Using Grad-CAM for SSD network

yizt / Grad-CAM.pytorch

pytorch实现Grad-CAM和Grad-CAM++,可以可视化任意分类网络的Class Activation Map (CAM)图,包括自定义的网络;同时也实现了目标检测faster r-cnn和retinanet两个网络的CAM图;欢迎试用、关注并反馈问题...

Apache License 2.0

721 stars 166 forks source link

Using Grad-CAM for SSD network #14

Open Fpooyan opened 4 years ago

Fpooyan commented 4 years ago

Hi, Thanks for publishing this great repo! I am debugging an SSD network used for a 3-class object detection task. Can I use this repo for visualizing such a network as well? I would appreciate it if you could provide me the initial steps...

yizt commented 4 years ago

@Fpooyan 您好，对于目标检测每个网络都不一样，SSD网络生成CAM图与RetinaNet比较类似，请问您基于哪个工程做SSD网络的可视化？可以互相交流一下~

Fpooyan commented 4 years ago

Thanks for your quick response! I do not speak Chinese, so I am using translate to get you. I want to find a way to visualize the base layers of my SSD so that I can see what features it is concentrating on. As SSD networks work with prior boxes, I thought their Grad-CAM must be something different.

yizt commented 4 years ago

@Fpooyan 您好，prior boxes 就是anchors；ssd,retinanet和faster r-cnn都是anchor-based 的目标检测；SSD和retinanet都是one-stage网络，都有特征金字塔，所有个人认为它们的Grad-CAM是类似的。请问你想基于哪个工程来做？