yolov3 not in model zoo?

[xiuyangleiasp]

❓ Questions and Help

[chenjoya]

Thanks for your question. YOLOv3 is not implemented on maskrcnn-benchmark. This repository only contains RetinaNet, Faster R-CNN, Mask R-CNN. Please wait.

[xiuyangleiasp]

Hi, @ChenJoya I was reproduced your "Optimal Bias Initialization" on yolov3-voc. It was amazing. Improved mAP about 1%. Thanks for your work. Would you release codes of your another paper “Residual Objectness for Imbalance Reduction”?

[chenjoya]

@xiuyangleiasp Thanks for your reproduction!

In fact, we did not try YOLOv3 with the sampling-free mechanism for PASCAL VOC. Would you like to describe your impeachmention details?

About the paper Residual Objectness for Imbalance Reduction, we have no plan to release the code as it submits to a journal now. The procedure of peer review is so long.

Again, thanks for your attempt!

[xiuyangleiasp]

the detail is very simple:

1. only init bias for last conv befor yolo;
2. init channels corresponding with objectness and classes.
3. I estimated N and Nf very roughly. N=feature_map size. Nf=1.25 for 26×26, and Nf=7.6 for 13×13 . C=20

And I have some questions about ResObj the pseudocode:

the case condition seems always be true. So I can't understand it's meanings. Would you explian it for me? Looking forward your reply. Thanks.

[chenjoya]

That sounds good.

Oh, 'minp' refers to the minimal positive score. We assume that negative examples whose scores below the 'minp' are learned well.

[xia00100]

thanks for author's work~

@xiuyangleiasp Hi~ would you release your code of YOLOv3 with Free-Sampling please, cause I want improve my work with this solution, but sadly I have no idea implement it.

Thank you honestly and looking forward your reply!

[xiuyangleiasp]

@xia00100 Sorry, I can't relase codes without authorization from the company.