Get Segmentation fault when set opset=9

[quanh1990]

Is the lowest version of onnx opset 10 ?

python3 tools/export_onnx.py \ --output-name weights/yolox_s.onnx \ -f exps/example/yolox_voc/yolox_voc_s.py \ -n yolox-s \ -c YOLOX_outputs/yolox_voc_s/best_ckpt.pth.tar \ -o 9

Then I got Segmentation fault (core dumped)

Thanks

[GOATmessi8]

Fine... If you really wanna use opset 9, you have to check out which operations this version doesn't support. I suppose the issue may come from the nearest upsample and the focus slice operator. So you may change to bilinear upsample (very little performance influence) and replace the focus with a normal conv 3x3 and fine-tune the model.

[quanh1990]

Fine... If you really wanna use opset 9, you have to check out which operations this version doesn't support. I suppose the issue may come from the nearest upsample and the focus slice operator. So you may change to bilinear upsample (very little performance influence) and replace the focus with a normal conv 3x3 and fine-tune the model.

Thanks a lot! I'll try.

[zhanggan123456]

Have you succeeded in visualization

[quanh1990]

Have you succeeded in visualization

Very successful, what is your problem?

[quanh1990]

Have you succeeded in visualization

没啥问题兄弟，你遇到啥问题？

[zhanggan123456]

你可视化成功了吗

没啥问题兄弟，你遇到啥问题？

可以pull一下你的执行语句吗？参数配置

[quanh1990]

你可视化成功了吗

没啥问题兄弟，你遇到啥问题？

可以pull一下你的执行语句吗？参数配置

哪个环节？训练/eval还是推理？

[zhanggan123456]

你可视化成功了吗

没啥问题兄弟，你遇到啥问题？

可以pull一下你的执行语句吗？参数配置

哪个环节？训练/eval还是推理？

网络结构可视化

[zhanggan123456]

你可视化成功了吗

没啥问题兄弟，你遇到啥问题？

可以pull一下你的执行语句吗？参数配置

哪个环节？训练/eval还是推理？

网络结构可视化

就是将pth权重文件转化为onnx格式文件的语句

[xiafan-su]

Hello, 请问你转onnx用opset=9成功了嘛，请问具体是怎么做的呢，有分支可以参考一下嘛？

[xiafan-su]

0001-Fix-onnx-export-opset-9-segmentation-fault-issue.patch.zip

尝试了个patch但是需要重新训练，感兴趣的可以试一下