enazoe / yolo-tensorrt

TensorRT8.Support Yolov5n,s,m,l,x .darknet -> tensorrt. Yolov4 Yolov3 use raw darknet *.weights and *.cfg fils. If the wrapper is useful to you,please Star it.
MIT License
1.18k stars 313 forks source link

tensorrt推理yolov5-v2模型,无法预测出结果 #82

Closed FanZhang91 closed 3 years ago

enazoe commented 3 years ago

https://github.com/enazoe/yolo-tensorrt/blob/master/modules/trt_utils.h#L217 https://github.com/enazoe/yolo-tensorrt/blob/master/modules/trt_utils.h#L221 把“hardswish”改成“leaky”

FanZhang91 commented 3 years ago

@enazoe 把“hardswish”改成“leaky”之后问题得到解决,棒棒的

alicera commented 3 years ago

请问有不用DLA 的方法去转INT8 吗? 我用一般方式去转 无法预测出结果

enazoe commented 3 years ago

请问有不用DLA 的方法去转INT8 吗? 我用一般方式去转 无法预测出结果

int8和dla没关系,你没结果也和dla没关系

zzcqinag commented 3 years ago

请问下,改成“leaky”之后还是出不来结果是什么原因呢?自己重新编译,engine重新生成的,python可以出结果 转之后的cfg: [net] width=640 height=640 channels=3 nc=1 depth_multiple=0.33 width_multiple=0.5 anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 是还有其他的地方要改吗?非常感谢

enazoe commented 3 years ago

请问下,改成“leaky”之后还是出不来结果是什么原因呢?自己重新编译,engine重新生成的,python可以出结果 转之后的cfg: [net] width=640 height=640 channels=3 nc=1 depth_multiple=0.33 width_multiple=0.5 anchors=10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326 是还有其他的地方要改吗?非常感谢

正常情况改完就没问题了,再不行我就不知道了,用我提供的2.0模型试吧

zzcqinag commented 3 years ago

就是用的2.0模型,是不是网络输入的尺寸不对? @enazoe

enazoe commented 3 years ago

@zzcqinag 你看对不对啊,肯定要改成你自己的啊,再不行看看检测阈值,把阈值调下去,你直接在群里说吧,