Part Number: AM67A
Hi,
I’m new to the EdgeAI workflow and I’m trying to compile an existing depth_estimation model from the TI Model Zoo for a specific TIDL version 11_00_06_00 targeting AM67A. I’m running into a few issues and would appreciate some guidance.
What I’m trying to do
I added the following custom model entry inedgeai-tidl-tools/examples/osrt_python/model_configs.py:
'de-7300' : create_model_config(
task_type="depth_estimation",
source=dict(
model_url="">software-dl.ti.com/.../fast-depth.onnx",
infer_shape=True,
),
preprocess=dict(
resize=256,
crop=224,
data_layout="NCHW",
resize_with_pad=False,
reverse_channels=False,
),
session=dict(
session_name="onnxrt",
model_path=os.path.join(models_base_path , "fast-depth.onnx"),
input_mean=[123.675, 116.28, 103.53],
input_scale=[0.017125, 0.017507, 0.017429],
input_optimization=True,
),
postprocess=dict(),
extra_info=dict(
num_images = numImages ,
num_classes = 1000
),
runtime_options = {
'advanced_options:inference_mode' : 2,
'advanced_options:num_cores' : 1,
'advanced_options:calibration_frames' : 12,
'advanced_options:calibration_iterations' : 12
}
)
Issue
The compilation completes, but insidemodel-artifacts/de-7300/artifacts/
no .bin files are generated. Instead, I only get a tempDir that contains temporary .bin files.
When I try to run those temporary bin files, they fail at runtime.
Additional problem
I also tried using edgeai-benchmark, but I couldn’t find a way to set the required TIDL version 11_00_06_00, which leads to version-mismatch errors.
Questions
-
What is the correct procedure to compile a Model Zoo model for a specific TIDL version (11_00_06_00)?
-
Is edgeai-benchmark expected to support manual TIDL version selection, or should it match the internal version of the tools repository?
Any guidance or examples would be very helpful.
Thank you!
