jkdxbns/autonomous-driving-carla

Autonomous Driving CARLA Models

Pre-trained models for autonomous driving in the CARLA 0.9.15 simulator.

Models

Model	File	Size	Description
YOLO11n	yolo_carla_best.pt	~6MB	Object detection (vehicles, pedestrians, traffic lights, speed signs)
UFLD	ufld_carla_best.pth	~735MB	Lane detection fine-tuned on CARLA
TuSimple	tusimple_18.pth	~245MB	Pre-trained lane detection backbone

YOLO11n Object Detection

Custom-trained on CARLA simulator data.

Classes:

• 0: Vehicle
• 1: Pedestrian
• 2: Traffic Light
• 3: Speed Limit Sign

Training:

• Epochs: 250
• Batch Size: 128
• Optimizer: AdamW
• Learning Rate: 0.005
• Image Size: 640×640

UFLD Lane Detection

Fine-tuned Ultra Fast Lane Detection model.

Architecture:

• Backbone: ResNet-18
• Grid: 100×56
• Input: 800×288

Training:

• Epochs: 50
• Batch Size: 32
• Optimizer: Adam
• Learning Rate: 1e-4
• Loss: SoftmaxFocalLoss + ParsingRelationLoss

Usage

from huggingface_hub import hf_hub_download

# Download YOLO model
yolo_path = hf_hub_download(
    repo_id="jkdxbns/autonomous-driving-carla",
    filename="yolo_carla_best.pt"
)

# Download UFLD model
ufld_path = hf_hub_download(
    repo_id="jkdxbns/autonomous-driving-carla",
    filename="ufld_carla_best.pth"
)

GitHub Repository

Full source code: github.com/jkdxbns/autonomous-driving-carla

Author

Justin Mascarenhas
CMPE 789 - Robot Perception
Rochester Institute of Technology

License

MIT License