this project is difficult to train on custom data

[futureflsl]

train on custom data document is too simple to follow,though the file exps/example/yolox_voc/yolox_voc_s.py can be trainning,but how the file relate with yolox_s? I saw the code,and I could not find keyword about yolox_s.
the document only support VOC,How about COCO data trainning?I hope witer can improve train on custom data skills and simplify the trainning step.Now I will give up this project.It is difficult to use!

[Joker316701882]

@futureflsl Thank you for your kind suggestion. We don't have many maintainers and we are now all working on more emergent issues and potential bugs. However, a more friendly tutorial is indeed in our plan for this week.

[zhangming8]

I analyzed the annotation format of COCO dataset and converted VOC data into COCO. You can refer to README.md here https://github.com/zhangming8/yolox-pytorch

[campos537]

train on custom data document is too simple to follow,though the file exps/example/yolox_voc/yolox_voc_s.py can be trainning,but how the file relate with yolox_s? I saw the code,and I could not find keyword about yolox_s.
the document only support VOC,How about COCO data trainning?I hope witer can improve train on custom data skills and simplify the trainning step.Now I will give up this project.It is difficult to use!

I think the def init is the part you need to change in order to choose other architecture

[wojiazaiyugang]

want to have a function that loads data. If you need to train custom data, you just need to override this function, like https://mmdetection.readthedocs.io/en/latest/tutorials/customize_dataset.html