dog-qiuqiu / Yolo-Fastest

:zap: Based on yolo's ultra-lightweight universal target detection algorithm, the calculation amount is only 250mflops, the ncnn model size is only 666kb, the Raspberry Pi 3b can run up to 15fps+, and the mobile terminal can run up to 178fps+
Other
2.01k stars 428 forks source link

请问,训练小模型时怎么对待极小人脸呢 #62

Open liuqunzhong opened 3 years ago

liuqunzhong commented 3 years ago

当训练小分辨率的模型时,在压缩到小分辨率的图像上,原本就比较小的人脸会变得极小,可能只有4、5个像素大小,请问怎么对待这些人脸呢? 1、如果设定比较小的anchor 4~10个像素的,这么小,感觉比较难训吧? 2、如果不想训练小于10个像素的,那应该是怎么对待呢,忽略(loss=0)还是当作负样本?