Training Details

[Gitlyz007]

Thanks for your job, I tried to run train.py, after configuring series of parameters , it seems the train code needs 'infoList.json' which is not exist in raw VOC2007 dataset. Could you please show more details of using the training code and the structure of your dataset?

[h1yuol]

The dataset I used was generated by myself. Here's the info: https://github.com/Yuol96/Siam-RPN/blob/0e10f13f52d3264f7e158e1f0162c16c9976e727/flag/builder.py#L204 You may write your own dataloader functions to adapt to your dataset.

Gitlyz007 notifications@github.com 于2018年11月11日周日 下午3:00写道：

Thanks for your job, I tried to run train.py, after configuring series of parameters , it seems the train code needs 'infoList.json' which is not exist in raw VOC2007 dataset. Could you please show more details of using the training code and the structure of your dataset?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Yuol96/Siam-RPN/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AjnY2GV4cTDVZxTFm0amMD_vsq_sC-Xcks5ut8sggaJpZM4YYWpy .

[wxh001qq]

@Yuol96 did you trained on VOC2007 dataset?

[h1yuol]

No. Sorry about that. So you may do your own data processing.

[QijinYin]

Hi!

I wander if there are multi targets for each image in your detection dataset, since there is not more than 1 target in each frame for tracking task, but usually several targets for detection task. There is "bboxList" in the "info" dict. Is this data structure suitable for tracking task ?

Thanks!

No. Sorry about that. So you may do your own data processing.

[h1yuol]

There are multiple targets for each image. I am not sure whether it's suitable for tracking, but it might be helpful to differentiate objects.

[QijinYin]

Thanks!