om-ai-lab/OmDet

onnx convert error

BaiCHao12 opened this issue · 3 comments

BaiCHao12 commented

onnx convert error
image

hx621 commented

@BaiCHao12 Thank you for your question. This issue was caused by torch.where， and we've made some changes to avoid this issue. Let me know if you have any other problems.

BaiCHao12 commented

language_backbone（model text embedding part） convert onnx error
image

hx621 commented

language_backbone（model text embedding part） convert onnx error image

@BaiCHao12 Due to the label cache mentioned in the paper, the language backbone only needs to be inferred once during model inference. Therefore, our project currently does not support the conversion of ONNX models with language backbones. If you have any other requirements, please let us know.