Stinky-Tofu
Stinky-Tofu
When testing mAP with the VOC2007 data set, should the bbox marked as "difficult" be removed?
Thanks, I didn't notice your script, and wrote one myself. After removing the bbox marked as "difficult", the mAP on voc2007 was upgraded from 81.5 to 85. https://github.com/Stinky-Tofu/YOLO_V3
I implement YOLO3 with tensorflow can only get 82% mAP without data enhancement and multi-scale training.
> @Stinky-Tofu ， Can you share your code? https://github.com/Stinky-Tofu/YOLO_V3
> 这个repo只有detection功能，那么怎么实现这个功能的呢？ > 写的一堆yolo model代码一点没用到，他只是用别人写好的yad2k.py文件把darknet.weight文件转成keras的.h5文件，然后load_model，就开始其实就是用训练好的模型，predict然后完事了。 > 所以我觉得没必要再看了，没有train功能，前面star的不知道真的看没看。 没有训练过的代码也能算得上复现吗？只有训练过，达到与论文相当的性能，才算复现出来了，自己复现的代码可能会与论文的性能相差甚远。达不到就说明存在很多问题，比如训练方式、数据增强、损失函数、标签等是否正确。
The optimal performance of my pedestrian detector(STNet will release) on cityPersons is 8% MR，and only need one GPU for train~
@juanmed Better performance, and provide YOLOV3-Lite
> Currently this model can't generate reasonable bounding boxes if training from scratch. There're several possible solutions > > 1. Training with enough epochs. According to [AlexeyAB/darknet](http://localhost:8890/#scalars&_ignoreYOutliers=false&runSelectionState=eyIwMm5kLjIyLjQzIjpmYWxzZX0%3D), in first 2000...