AlphaPose icon indicating copy to clipboard operation
AlphaPose copied to clipboard

请问能分享一下Halpe136关键点的预训练模型的训练过程吗?

Open hongyaohongyao opened this issue 3 years ago • 6 comments

@Fang-Haoshu 您好,我现在正在尝试训练更轻量的SPPE。我现在想知道Halpe136关键点的预训练模型是不是并不只是在那个Halpe项目上提供的数据集上进行了训练,而是用了很多其他的数据集做了预训练。请问是这样吗?如果是,能否分享一下训练的过程,比如依次用了哪些数据集,配置了什么样的yaml文件?非常感谢。

hongyaohongyao avatar Jun 05 '21 10:06 hongyaohongyao

@Fang-Haoshu I trained alphapose in halpe 136 dataset, but the MAP is worse than the pretrained_model(halpe136_fast_res50_256x192.path),Are there any details to pay attention to?and could you shard the train log ?

abcliguanxi avatar Jun 07 '21 07:06 abcliguanxi

@hongyaohongyao ,您好,训练halpe136的预训练模型有吗。能否提供一下下载地址。还有halpe136的数据集与coco是否相关

fairleehu avatar Jun 16 '21 01:06 fairleehu

@hongyaohongyao ,您好,训练halpe136的预训练模型有吗。能否提供一下下载地址。还有halpe136的数据集与coco是否相关

预训练模型在他们的项目里有提供model_zoo,第二个问题我也不清楚,看起来像是coco 2015年的那个版本的重新标注

hongyaohongyao avatar Jun 16 '21 15:06 hongyaohongyao

@hongyaohongyao ,您好,训练halpe136的预训练模型有吗。能否提供一下下载地址。还有halpe136的数据集与coco是否相关

预训练模型在他们的项目里有提供model_zoo,第二个问题我也不清楚,看起来像是coco 2015年的那个版本的重新标注 好的,谢谢

fairleehu avatar Jun 18 '21 05:06 fairleehu

Hello, have you solved it now? The MAP I trained with HALPE26 was low and overfitting. Is it possible that I trained with the HALPE dataset and COCO together?

xiaochenwise avatar Jun 22 '21 02:06 xiaochenwise

@Fang-Haoshu I trained alphapose in halpe 136 dataset, but the MAP is worse than the pretrained_model(halpe136_fast_res50_256x192.path),Are there any details to pay attention to?and could you shard the train log ?

May I ask if you have solved it? Such problem also appeared in my training with HALPE26 data set.

xiaochenwise avatar Jun 22 '21 08:06 xiaochenwise