boyang9602/tlr-replicate

ApolloTLR Python Replication

This is a python / pytorch replication of Apollo Traffic Light Detection and Recognition (TLR).

You can install it by python3 -m pip install -r requirements.txt --user . You may need to install and configure pytorch with cuda.

Example code

from tlr.pipeline import load_pipeline
import cv2
tlr = load_pipeline() # load_pipeline('cuda:0)
valid_detections, recognitions, assignments, invalid_detections = tlr(image, projection_bboxes)

Output

1. valid_detections is a n * 9 tensor. The first column is useless in this project. 1:5 are the bounding boxes, 5:9 are the TL type scores vector.
2. recognitions are the recognition scores vector.
3. assignments is a n * 2 tensor. Each row is match between the projection and the valid detection. The first col is the idx of a projection of TLs and the second col is the idx of a valid detection.
4. invalid_detections are discarded in Apollo.

Please refer to https://github.com/ApolloAuto/apollo/blob/v7.0.0/docs/specs/traffic_light.md for a high-level understanding of Apollo TLR.

This project is part of SITAR: Evaluating the Adversarial Robustness of Traffic Light Recognition in Level-4 Autonomous Driving. Please consider to cite it if you found it useful.

@inproceedings{sitar,
author={Yang, Bo and Yang, Jinqiu},
title={SITAR: Evaluating the Adversarial Robustness of Traffic Light Recognition in Level-4 Autonomous Driving},
year={2024},
booktitle={35th IEEE Intelligent Vehicles Symposium}
}