search

aim-uofa / AdelaiDet

AdelaiDet is an open source toolbox for multiple instance-level detection and recognition tasks.
https://git.io/AdelaiDet
Other
3.37k stars 646 forks source link

Question of exporting ONNX on CondInst. #399

Open xiehousen opened 3 years ago

xiehousen commented 3 years ago

@tianzhi0549 Hello, When I use the file onnx/export_model_to_onnx.py encountered a question I don't understand.

When outputting the CondInst onnx, dynamic_mask_head part used as post-processing? Why is there no dynamic_mask_head in the output ONNX?

So, If there is no dynamic_mask_head part in onnx, do we need to add dynamic_mask_head in post-processing?

Thanks a lot for your reply.

JJC6050 commented 2 years ago

When I export boxinst to onnx, I encounter RuntimeError: ArrayRef: invalid index Index = 21; Length = 21.My instructions are:python onnx/export_model_to_onnx.py --config-file configs/BoxInst/MS_R_50_3x.yaml --output /data/pretrained/onnx/BoxInst/BoxInst_R_50_3x_bn_head.onnx --opts MODEL.WEIGHTS training_dir/BoxInst_MS_R_50_3x/model_final.pth Can you tell me what's wrong？

605436079 commented 2 years ago

@tianzhi0549 Hello, When I use the file onnx/export_model_to_onnx.py encountered a question I don't understand.

When outputting the CondInst onnx, dynamic_mask_head part used as post-processing? Why is there no dynamic_mask_head in the output ONNX?

So, If there is no dynamic_mask_head part in onnx, do we need to add dynamic_mask_head in post-processing?

Thanks a lot for your reply.

Has the problem been solved? I also found that this network layer is difficult to write after it is put in place, and the speed is slow