hyper-paramter "fuse_ab"

[QiqLiang]

Before Asking

• I have read the README carefully. 我已经仔细阅读了README上的操作指引。

• I want to train my custom dataset, and I have read the tutorials for training your custom data carefully and organize my dataset correctly; (FYI: We recommand you to apply the config files of xx_finetune.py.) 我想训练自定义数据集，我已经仔细阅读了训练自定义数据的教程，以及按照正确的目录结构存放数据集。（FYI: 我们推荐使用xx_finetune.py等配置文件训练自定义数据集。）

• I have pulled the latest code of main branch to run again and the problem still existed. 我已经拉取了主分支上最新的代码，重新运行之后，问题仍不能解决。

Search before asking

• I have searched the YOLOv6 issues and found no similar questions.

Question

Hi, i want to know if fuse_ab=True means that we use anchors-based branch as auxiliary when training while we throw it away when infering? Cuz i print the model structure while infering and find that there is still a cls_preds_ab branch and a reg_preds_ab branch printed. Is it more reasonable that there is only cls_preds_af branch and reg_preds_af branch？Why does it happen and how can i solve it?

Additional

No response

[Chilicyy]

@QiqLiang Hi, if applying fuse_ab mode, the cls_preds_ab and reg_preds_ab branch are thrown away when you export onnx model for later application. When you run inference with pytorch checkpoint model, it would only use the predictions from anchor free branches, while the anchor-based branches can still be found in checkpoint model.

[QiqLiang]

Thanks for answering! Could you please further supply the code link to show where they are thrown away? im kind of confused about it and want to see the corresponding code, thanks again!

[Chilicyy]

Hi, you can refer to this line.

[roxroxroxrox]

after turn on this params , the training time doubled.
cannot afford this cost...