How can I get predicted class in eval.py

[atinfinity]

In https://github.com/AI-liu/Complex-YOLO#result, predicted box and predicted class is drawn.
But, I think that eval.py draws only predicted box and target box. So, I want to know how to get predicted class from all_boxes.

[abhigoku10]

@atinfinity i am too facing the same issue did u get a solution for it

[atinfinity]

@abhigoku10
I think that pred_boxes has 15 elements. And, pred_boxes[7]~[14] has the confidence of classes.
https://github.com/AI-liu/Complex-YOLO/blob/master/eval.py#L59-L68

But, some element has negative value. So, I don't understand this reason.

[abhigoku10]

@atinfinity thanks for the inputs, yes negatives even i observed it and had one query does the prediction of the classes are proper

[abhigoku10]

@atinfinity where you able to get correct prediction of the classes . even i am getting negative values in the tensors

[abhigoku10]

@atinfinity i was able to solve the issue let me know if you have not solved it .

[siliconvalleypilot]

@abhigoku10 How did you solve the problem? I see negative values and np.argmax does not give the correct label position

[abhigoku10]

@qiaosongwang please take the updated code from the repo and train it you shall get correct values or i have forked the repo and made the corrections you can also take it fromthere

[siliconvalleypilot]

@abhigoku10 Great! Thanks

[imdsafi09]

I train the model till 158 epochs and then eval on different epoch but the detection is only for car which is not correct enough and the results here contain the class labels also but in my case no labels either..can someone please guide me whats the problem

[abhigoku10]

@imdsafi09 i would suggest you look into https://github.com/ghimiredhikura/Complex-YOLOv3 this is currently updated and the current repo is not by the author and is not updated also

[imdsafi09]

@abhigoku10 thanks for your reply. I am already working on your mentioned repo but i want to use it with vlp-16 and that case i dont know how to train it for my own data because the KITTI data set you know is based on vlp-64