Issue Compiling Optimized HAR to HEF for Lane Detection Model (CULane Dataset)

General
, , ,
1

Hi everyone,

I’m working on optimizing and compiling the resnet18 ONNX model from the Vehicle-CV-ADAS GitHub repository, which is trained on the CULane dataset. This model is based on the Ultra-Fast Lane Detection v2 architecture.

So far, I’ve completed the following steps:

  • Successfully parsed the ONNX model into a HAR file.
  • Used 1032 images from the CULane dataset to optimize the HAR file.

However, I’m encountering an error when trying to compile the optimized HAR into an HEF file.

My questions:

  1. What kind of dataset is required during the optimization/compilation process?
  2. Do I need to provide any additional annotations (e.g., lane markings, segmentation masks) along with the images?
  3. Is there any specific preprocessing or format I should follow for the dataset used in this step?

Any guidance or suggestions on how to resolve this compilation issue or prepare the dataset correctly would be greatly appreciated.

Thanks in advance!

2

Hi @omria @Klausk,
We’re facing the same issue mentioned above during the HEF compilation step after optimizing the HAR file. Could you please help us out with this? Appreciate your support!