Hi,
I use the official edgeai-yolov5s model for importer, Encountered the following problems.
1.InElementType=0 in the configuration file tidl_import_xx.txt.Execute the command: ./out/tidl_model_import.out /../ti_dl/test/testvecs/config/import/public/onnx/tidl_import.txt In the terminal window, output elementType=6 for the first time. At this time, the output test picture can correctly detect the target. Then the terminal window outputs elementType=6 again, at this time the output test picture cannot detect the target correctly. Query the document to know that elementType is a supported quantitative type. Why are the output results of the same type of quantization operation very different?
2.A semantic segmentation branch is added on the basis of the yolov5s model. Does the modified model support model quantization operations?
3.Add a semantic segmentation branch based on the yolov5s model .Execute the command ./out/tidl_model_import.out /../ti_dl/test/testvecs/config/import/public/onnx/tidl_import_xx.txt, the first detection branch and semantic segmentation branch output in the terminal window are elementType=6 At this time, the output test image can correctly detect the target but the semantic segmentation result is incorrectly segmented. Then the terminal window once again outputs element Type=0 in the semantic branch and elementType=6 in the detection branch. At this time, the output test image cannot detect the target correctly and there is no semantic segmentation result. Is this what is the reason? Part of the terminal output is as follows:
# 0 . .. T 3806.62 .... ..... ...
out[]->elementType 6
test tidl_NetOutputPostproc ioPrms->numOutputBuf 0
test tidl_NetOutputPostproc ioPrms->outWidth 384
test tidl_NetOutputPostproc ioPrms->outHeight 384
test tidl_NetOutputPostproc ioPrms->outNumChannels 2
out[]->elementType 6
test tidl_NetOutputPostproc ioPrms->numOutputBuf 1
test tidl_NetOutputPostproc ioPrms->outWidth 1404
test tidl_NetOutputPostproc ioPrms->outHeight 1
test tidl_NetOutputPostproc ioPrms->outNumChannels 1
.... .....
~~~~~Running TIDL in PC emulation mode to collect Activations range for each layer~~~~~
----------------------- TIDL Process with REF_ONLY FLOW ------------------------
# 0 . .. T 2706.76 .... ..... ...
out[]->elementType 0
test tidl_NetOutputPostproc ioPrms->numOutputBuf 0
test tidl_NetOutputPostproc ioPrms->outWidth 384
test tidl_NetOutputPostproc ioPrms->outHeight 384
test tidl_NetOutputPostproc ioPrms->outNumChannels 2
out[]->elementType 6
test tidl_NetOutputPostproc ioPrms->numOutputBuf 1
test tidl_NetOutputPostproc ioPrms->outWidth 1404
test tidl_NetOutputPostproc ioPrms->outHeight 1
test tidl_NetOutputPostproc ioPrms->outNumChannels 1
Thanks in advance,
Cheers