Only detection?
Opened this issue · 6 comments
Are you only detect the object but not classify it?
I would like to know about this as well, both this and the other ComplexYOLO code seem to not output any class predictions. From the code it seems they are included (output[7:] seems to be the magic location), but due to my unfamiliarity with both YOLO and PyTorch I simply cannot crack how I'm supposed to get a meaningful prediction.
Yes, I figured it out. Basically the output matrix you get, the last columns are the class scores, i.e.
they are assembled like this: pred_boxes[..., 7:(7 + nC) ] = pred_cls.
So the last columns from the method get_region_boxes(output, conf_thresh, num_classes, anchors, num_anchors) are what you are looking for.
@Artamus thanks for the pointers, i was able to get the class label . The issue which i am facing now is that their misclassifcation in the class label.? are you also facing the same issue. I have tried to use sigmoid and softmax both are yielding same result
If you just want to get the highest scored class, argmax should be enough (i.e. just get the class with highest score, no need to pass it to a softmax).
As to the accuracy, sorry, I cannot help you there, I only did a quick evaluation of ComplexYOLO and I am not using it anywhere.
@Artamus Hi, you can take a look at this project, which has a complete implementation and effect presentationhttps://github.com/wwooo/tensorflow_complex_yolo