tinyvision / DAMO-YOLO

DAMO-YOLO: a fast and accurate object detection method with some new techs, including NAS backbones, efficient RepGFPN, ZeroHead, AlignedOTA, and distillation enhancement.
Apache License 2.0
3.78k stars 476 forks source link

readme里的"Finetune on your data"是指什么? #62

Closed fppccc closed 1 year ago

fppccc commented 1 year ago

Before Asking

Search before asking

Question

您好!请问"Finetune on your data"是指训练吗? 我已经按照[custom dataset tutorial] (https://github.com/tinyvision/DAMO-YOLO/blob/master/assets/CustomDatasetTutorial.md) 在自己的小数据集上尝试了,每一步都按照教程来,使用了M模型,并准备好了预训练权重。但是AP值几乎为0,这个小数据集之前在其他模型上跑过,可以确定格式没问题(coco格式)。请问我该如何解决问题呢? 非常感谢!

Additional

No response

cwhgn commented 1 year ago

请问下您的数据集train和test的数据集类别和标签是一致的么?

XianzheXu commented 1 year ago

@fppccc 麻烦看一下您数据集的train.json和val.json中的categories字段中的类别映射排序是否一致,在当前的版本中,若train.json和val.json中的categories字段中的类别映射顺序不一样,会导致train和val时的类别映射不对齐。您可以手动将train.json和val.json中的categories字段中的类别映射顺序调整一致来避免该问题。后续我们会推出新版本修复该问题,敬请保持关注。

HYLcool commented 1 year ago

This issue will be closed due to inactivity. Please reopen as needed.