DetectionOutputLayer

[GraphicsHunter]

Hi. I tried to implement the detection output layer as you suggested in the previous issue. But when I run my implementation of that layer I always get 200(which is what my keep_top_k value is set to) detections, even when there are no objects on the screen. If I look at the number of detections with Caffe, I see that the numbers are dynamic and depends on if the object is on the screen or not. Do you also have the same "problem" with your implementation?

[chenzhi1992]

Top_k doesn't represent the number of detections with caffe and the final output. You need to look at the code again, and remember to verify that the data of every layer is the same as caffe's output.

[cp625128752]

I get success when I run zf-ssd (300x300),but run resnet34-ssd (330x330) was wrong.Detection_output have the same bottoms with caffe.And detection_output's parameters all is right. I suspect the detection_output layer is a problem.

[cp625128752]

@chenzhi1992