CD543 / ECE549 Final Project — Attention-Guided Lane Detection

This repository contains our final project for CD543/ECE549 (Computer Vision).
We re-implement a simplified version of LaneATT to study its architecture, hyperparameters, and performance on small subsets of TuSimple or CULane datasets.

---

Setup

1. Install Poetry 1.1.12

curl -sSL https://install.python-poetry.org | POETRY_VERSION=1.1.12 python3 -
export PATH="$HOME/.local/bin:$PATH"
poetry --version

2. Clone the repository

git clone [email protected]:YOURUSER/laneatt-classproj.git
cd laneatt-classproj

3. Initialize the submodule

Run these two commands once after cloning:

git submodule init
git submodule update

That will:

• Register the submodule
• Check out the correct commit inside external/LaneATT

You should then see the LaneATT code under external/LaneATT/.

4. Install dependencies and activate the environment

poetry install --no-root
poetry shell

5. Verify installation

python -c "import torch, torchvision, numpy, cv2; print(torch.__version__, torchvision.__version__, numpy.__version__, cv2.__version__)"

If version numbers print without errors, the setup is complete.

Optional: GPU Setup

If you have an NVIDIA GPU:

pip install --upgrade --index-url https://download.pytorch.org/whl/cu121 torch==2.5.1 torchvision==0.20.1
python -c "import torch; print(torch.__version__, torch.cuda.is_available())"

Notes

• Requires Python 3.10–3.14
• Tested with Poetry 1.1.12 and PyTorch 2.5.1
• For reproducibility, all dependencies are defined in pyproject.toml and poetry.lock