SyGoing

I am looking forward to your work too. It's so good .If possible, could you please send the ultra-high precision model to [email protected]? Thank you！

the question is the title

关于训练的log.txt作者能提供下吗？

1

您好，非常感谢的你的工作，我想请问一下一般训练的总loss(class+loc)为多少的时候才算收敛呢？ 或者location loss为多少才合适呢

Could you please provide the C5's DataSet python file?

1

Since there is only the gen_dataset_visdrone.py , how can create the C5's dataSets

Could you please provide the shell that convert Centernet/Centernet2 to onnx ?

4

Since the model is trained to deplay on the application, is there any difference between detectron2 's default model export and centernet2 or centernet*

我用你的代码在yolov2稀疏化训练时（yolov2 608输入，两类目标）， 1）报错提示： 损失非常大，然后崩了，梯度爆炸。还有就是utils.py里的build_targets 提示index=19 in dimension 3 超过边界。 2）源码yolomodel.py中 self.losses["recall"] /= 3 self.losses["precision"] /= 3 为什么要除于3，是因为yolov3有三个输出层么？ 那么yolov2的话是不是需要修改呢

because of the Internet problem,I cannot download your pretrained model, could you please give the baidu yun downloads?

I found that your code have no operation about the thranspose the img, did you train your model without transpose the model?

What the caffe2's inference cost time?

1

You do a good job.Since you have idea to make the detection faster, can you show some results about the inference time? Thanks.