Open noringname opened 2 years ago
This error is a bit surprising because this model is in our modelzoo: https://github.com/TexasInstruments/edgeai-modelzoo/blob/master/models/vision/detection/coco/edgeai-mmdet/ssd_regnetx-200mf_fpn_bgr_lite_320x320_20201010_model.onnx.link https://github.com/TexasInstruments/edgeai-modelzoo/blob/master/models/vision/detection/coco/edgeai-mmdet/ssd_regnetx-200mf_fpn_bgr_lite_320x320_20201010_model.prototxt.link
And we have compiled it recently and the compiled artifact is here: https://github.com/TexasInstruments/edgeai-modelzoo/blob/master/modelartifacts/8bits/od-8050_onnxrt_coco_edgeai-mmdet_ssd_regnetx-800mf_fpn_bgr_lite_512x512_20200919_model_onnx.tar.gz.link
Hi @mathmanu, Thank you for your reply, I noticed that in the prototxt file you provide, there are also two output fields, so it will encounter the same error. I am wondering if this has to do with the SDK version (I am using 7.3). If so, what training framework should I use? I remember there was a pytorch-mmdetection. Thanks.
Hi @mathmanu, for some reason, we need to stick to SDK7.3, and I cannot get the models in this repo working, are there any repo I can use to train and deploy OD models to SDK7.3?
Hi, can you try an older branch of this repository and see if it helps?
Hi @mathmanu I tried r8.0, still not working, for the yolov3_regnet, ssd_regnet examples in SDK7.3, where can I find the training framework? Other than that, If I just want to import a custom OD model, I don't know how to write the meta.prototxt, please help.
r8.0 is close enough for SDK7.3 If the trained/exported model doesn't work, you can file an e2e issue to help from our TIDL experts.
Hi, I trained a ssd_regnetx_fpn_bgr_lite model and encountered some problems while importing it to bin.
Thanks.