Blitznet网络理解笔记
1 resnet的理解
https://blog.youkuaiyun.com/lanran2/article/details/79057994
2 DSSD的理解
https://www.cnblogs.com/xuanyuyt/p/7447111.html
https://blog.youkuaiyun.com/jesse_mx/article/details/55212179
3 Bliznet网络
https://blog.youkuaiyun.com/yaoqi_isee/article/details/77195029
https://blog.youkuaiyun.com/zhangjunhit/article/details/77577076
https://blog.youkuaiyun.com/shenxiaolu1984/article/details/51428392
4 Non-Maximum Suppression,NMS
https://www.cnblogs.com/makefile/p/nms.html
5 mAP理解
https://blog.youkuaiyun.com/zdh2010xyz/article/details/54293298
一、Introduction
BlitzNet: A Real-Time Deep Network for Scene Understanding
目标是做到实时的目标检测和语义分割,通过的简单的全卷积网络解决两个任务.在论文中作者表示两个任务的大部分权值是共享的.最终可以证明两个任务是一种相互促进的关系.
论文中作者给出的结果:
推理硬件:Titan X(Maxwell)
在Pascal Voc数据集上表现如下,300表示推理时的图片尺寸是300*300
model | FPS | mIoU | mAp |
---|---|---|---|
BlitzNet300 | 24 |