Tool/software:
We are encountering several issues with the AM62A 9.2 TIDL toolchain that require support:
1. Bin Inference:
When exporting the bin in the X86 Container, using PCxxx.out for inference results in a Segment Fault. We need this method for X86 simulation to integrate into our toolchain for evaluation.
2. Abnormal Accuracy with QDQ Model:
During the QDQ process with the edgeai-tidl tool, we compared output features of the ONNX and bin after QDQ and found discrepancies. Do you have any demos or guidance on this?
Parameter Configuration:
Compilation Errors:
While compiling mobilenet_v2_lite_wt-v2_qat-v2-wc8-at8_20231120_model.onnx, the following errors were reported:
Inference Results:
After simplifying the ONNX model(with onnxsim), compilation succeeded, but precision was misaligned:
Inference Image:：airshow.jpg
Inference QDQ ONNX: 0 17.885122 warplane, military plane
Inference bin: 0 18.246786 warplane, military plane
We would appreciate any assistance or guidance on these issues.
