---

name: model-export description: "Routes LaneNet checkpoint freezing and optional MNN/mobile export workflows for frozen PB and MNN deployment." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Model Export

Use this sub-skill when the user has a LaneNet TensorFlow checkpoint and wants a frozen .pb graph, or when they need to understand the optional .pb to MNN/mobile deployment path.

Route First

• Need to create or resume checkpoints? Read ../training/SKILL.md first.
• Need to run checkpoint-backed prediction or TuSimple evaluation before export? Read ../inference-evaluation/SKILL.md first.
• Need frozen PB or MNN/mobile export details? Read references/mnn-export.md.
• Export failed or the MNN runtime cannot find tensors/config values? Read references/troubleshooting.md.

Core Helper

The bundled freeze helper is scripts/freeze_lanenet_model.py. It preserves the repository freeze API convert_ckpt_into_pb_file(ckpt_file_path, pb_file_path) and adds CLI help, repo-root handling, checkpoint preflight checks, and output-node reporting.

Quick shape from this sub-skill directory:

python scripts/freeze_lanenet_model.py \
  --repo-root <lanenet-repo-root> \
  --weights_path <checkpoint-prefix> \
  --save_path <output.pb>

The frozen graph uses these fixed nodes:

• input: lanenet/input_tensor
• binary segmentation output: lanenet/final_binary_output
• pixel embedding output: lanenet/final_pixel_embedding_output

Boundaries

This sub-skill documents checkpoint freeze, MNN converter command shape, MNN config.ini fields, runtime preprocessing expectations, and export-specific troubleshooting. It intentionally does not bundle the MNN converter shell script or document a full C++/MNN build, because the converter and mobile runtime toolchain are external to the Python LaneNet repository.