Bubbliiiing

Sorry, I don't know. I make this code from keras, it support generator better.

emm VOC数据集么？你可以下下来算一下，正好熟悉一下过程

我觉得应该是你打开的根目录错误了

肯定用了啊

You can refer to this code to generate the txt required for training coco. https://github.com/qqwweee/keras-yolo3 I don't have a powerful machine to complete coco training and eval

这个loss要降到0.几，你的图片很少？

增大训练集再说，我做过归一化了兄弟

Emmmmm我搞错了。。我的问题，重新上传了

emm？好像还是挺大的吧，建议用yolo4tiny

是