This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

BEAGLEY-AI: TI DL Model compilation issue - compiler crash

Part Number: BEAGLEY-AI

Tool/software:

General info
------------
Model: 1.onnx (exported from yolov8n, ONNX opset 12, onnx==1.13.0)
Compile scripts:
compile_model.sh // <- run this
compile_model.py
Complete console logs of model compilation:
with debug_level=1: log1.txt
with debug_level=3: log3.txt
Model works with TIDL offload disabled (when using only CPUExecutionProvider).

Issue
-----
Looks like there is some bug in Conv node processing code.
(see crash_location1.txt):

1. It crashes in tidlConstraintConv code
2. It crashes after the following lines are printed, so successfully processed nodes are Conv, Sigmoid, Mul. The next node to process is '/model.1/conv/Conv' (see netron_1_onnx.png):

SUPPORTED: Layers type supported by TIDL --- layer type - Conv, Node name - /model.0/conv/Conv
SUPPORTED: Layers type supported by TIDL --- layer type - Sigmoid, Node name - /model.0/act/Sigmoid
SUPPORTED: Layers type supported by TIDL --- layer type - Mul, Node name - /model.0/act/Mul


3. It doesn't crash in that location, if the following is added to compilation script (but it fails in another place, see 'Another error' section):
ti_options['deny_list:layer_type'] = 'Conv'

Another error
-------------
If Conv layer type is in deny list, then there is another error.
Complete console output in this case (debug_level=6): log6_NO_Conv.txt

Error:
Non-zero status code returned while running Conv node. Name:'/model.1/conv/Conv' Status Message: X num_dims does not match W num_dims. X: {1,1,1,16,320,320} W: {32,16,3,3}
and crash in different location (see crash_location2.txt)
compile_model_issue.zip