Test detection error

[gjm441]

I have trained the model as instructions, and run the code for testing VOC image, but it has poor result. After I learn the darknet yolo v2 source code, I find that the box_data_layer may be wrong, because it still takes side_side_(1+1+1+4), the same with yolo v1, that means one cell still only have one label class.

I have modified the box_data_layer implementation according to yolo v2 based on darknet. But it has some link problems.

[karta0807913]

oops, you're right! I will try to fix that The reason I did not change before is that I might to rewrite yolo v1 maybe I can add a new layer?

[zhengstake]

I have just started looking at this code base. I'd want to make sure that it is got the basic YOLO V2 function implemented correctly. I am a little bit confused by this issue and the answer.

I thought the box_data_layer was just importing the ground truth labels and output a top_blob[1] = 1313(1+1+1+4), which for each cell contains 'difficult, isobj flag, class label and 4 box parameters. This should have nothing to do with the per anchor box predictions. This ground truth label blob also seems to be used consistently in the yolo_v2_loss layer.

So what exactly is the expected changes to be made for YOLO V2?

[karta0807913]

I just change the loss function base this implement so that part is yolov1 and i didn't change that