mkoehnke/carnd-capstone-trafficlightdetection

Traffic Light Detection with Tensorflow Object Detection API

This repository was used to setup an environment to train/evaluate a pre-trained model to identify/classify traffic lights in pictures using tensorflow. It was part of my final project for the Udacity Self-Driving Car Engineer Nano Degree program.

Setup

Dependencies

sudo apt-get install protobuf-compiler
sudo pip install pillow
sudo pip install lxml
sudo pip install jupyter
sudo pip install matplotlib

Protobuf Compilation

# From root directory
protoc object_detection/protos/*.proto --python_out=.

Add Libraries to PYTHONPATH

# From root directory
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

Download SSD MobileNet Model

# From root directory
curl http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz | tar -xv -C model/ --strip 1

Creating TFRecord files

python object_detection/dataset_tools/create_pascal_tf_record.py --data_dir=data/sim_training_data/sim_data_capture --output_path=sim_data.record

python object_detection/dataset_tools/create_pascal_tf_record.py --data_dir=data/real_training_data/real_data_capture --output_path=real_data.record

Training

For Simulator Data

Training

python object_detection/train.py --pipeline_config_path=config/ssd_mobilenet_v1_coco_sim.config --train_dir=data/sim_training_data/sim_data_capture

Saving for Inference

python object_detection/export_inference_graph.py --pipeline_config_path=config/ssd_mobilenet_v1_coco_sim.config --trained_checkpoint_prefix=data/sim_training_data/sim_data_capture/model.ckpt-6000 --output_directory=model_frozen_sim/

For Real Data

Training

python object_detection/train.py --pipeline_config_path=config/ssd_mobilenet_v1_coco_real.config --train_dir=data/real_training_data

Saving for Inference

python object_detection/export_inference_graph.py --pipeline_config_path=config/ssd_mobilenet_v1_coco_real.config --trained_checkpoint_prefix=data/real_training_data/real_data_capture/model.ckpt-25000 --output_directory=model_frozen_real/

Attributions

• https://becominghuman.ai/traffic-light-detection-tensorflow-api-c75fdbadac62

carnd-capstone-trafficlightdetection's Issues

about Creating TFRecord files

When I use this command to create TFRecord, the sim_data.record is empty.
python object_detection/dataset_tools/create_pascal_tf_record.py --data_dir=data/sim_training_data/sim_data_capture --output_path=sim_data.record

I found there is some code in the "create_pascal_tf_record.py"
files = glob.glob(os.path.join(data_dir, "*.xml"))

So, it there some xml file missing in /data/sim_training_data/sim_data_capture/ ?

Thanks a lot