quic / qidk

Other
93 stars 18 forks source link

YOLOv8 quantized not running on DSP or GPU Runtime #36

Open muhammad-qasim-cowlar opened 1 day ago

muhammad-qasim-cowlar commented 1 day ago

SDK Version: 2.26 I can successfully execute snpe-net-run on my cpu and gpu on the unquantized model. I am following all the steps in this repository and I identified two layers to add to the .encodings file however when using this quantized model on dsp I get the following output:

snpe-net-run --container yolo_quantized.dlc --input_list list.txt --output_dir $OUTPUT_FOLDER --use_dsp --debug    
error_code=1002; error_message=Layer parameter value is invalid. No backend could validate Op=/model.22/Sigmoid_1 Type=ElementWiseNeuron error code=3110; error_component=Model Validation; line_no=142; thread_id=546085487872

The device I am using is sm8250 with v66 DSP arch. I have no idea what is causing this error.

quic-vraidu commented 16 hours ago

QIDK project is validated with SM8550 and SM8650. I recommended to use these devices.