建议导出的onnx输出层，不管是S模型、M模型还是T模型，导出层的名称保持一致，要不然解析需要做分支

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

3.75k stars 470 forks source link

建议导出的onnx输出层，不管是S模型、M模型还是T模型，导出层的名称保持一致，要不然解析需要做分支 #60

Closed xinsuinizhuan closed 1 year ago

xinsuinizhuan commented 1 year ago

Search before asking

[X] I have searched the DAMO-YOLO issues and found no similar feature requests.

Description

T模型的onnx： S模型的onnx: Mmo模型的onnx:

Use case

No response

Additional

不同的模型，我得重新写或者做分支区分类别来对应不同的解析，建议整成一样的。

Are you willing to submit a PR?

[X] Yes I'd like to help by submitting a PR!

cwhgn commented 1 year ago

感谢建议，后面我们新版本中会尽量保证一致。