hexiangquan / Traffic_Light_Localization_Recognition

Traffic_Light_Localization_Recognition (WIP Repo)

How to Fine-Tune:

1. Install TF object detection API https://github.com/tensorflow/models/tree/master/research/object_detection

2. Download ssd_mobilenet_v1_coco https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

3. Download Bosch traffic light dataset https://hci.iwr.uni-heidelberg.de/node/6132

4. Create following folders: checkpoints, data, input, model

Training Command

python train.py --pipeline_config_path=ssd_mobilenet_v1_coco.config --train_dir=./checkpoints

Inference Graph Builder

python export_inference_graph.py \ --input_type image_tensor \ --pipeline_config_path ./ssd_mobilenet_v1_coco.config \ --trained_checkpoint_prefix ./checkpoints/model.ckpt-<check_point_number> \ --output_directory ./model