- The objective is to create an Android Application to aid visually challenged people in navigation by detecting common obstacles on the street and notifying the user about the same.
- This project was developed as a part of the coursework CS256-Topics in Artificial Intelligence under the supervision of Dr. Natalia Khuri at San Jose State University during Aug-Dec 2017.
- The team members include Shantanu Deshmukh and Saketh Saxena.
- Install Docker Toolbox
- Open Docker QuickStart Terminal
- To create a new docker container:
docker run -it gcr.io/tensorflow/tensorflow:latest-devel - CHECK CONTAINER NAME:
docker ps -a - To enter this container:
docker start CONTAINERNAME
docker attach CONTAINERNAME
CONTAINERNAME = kind_fermat for this example
-
COPY FILES FROM WINDOWS TO DOCKER: (all files within the folder are copied, not the folder)
docker cp F:/#SJSU/AI/PROJECT/Training kind_fermat:/aiproject/Tests -
COPY FILES FROM DOCKER TO WINDOWS:
docker cp kind_fermat:/aiproject/Tests F:/#SJSU/AI/PROJECT/Testing -
Insde the container, execute the following commands:
#: cd tensorflow
#: git pull
IMAGE_SIZE=224
ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"
python tensorflow/tensorflow/examples/image_retraining/retrain.py \
--bottleneck_dir=/aiproject/bottlenecks \
--model_dir=/aiproject/inception \
--output_graph=/aiproject/retrained_graph.pb \
--output_labels=/aiproject/output_labels.txt \
--image_dir /aiproject/Train \
--architecture="${ARCHITECTURE}" \
--summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}"
python <script> <image_path> <label_file.txt> <graph_file.pb>
- Example:
python /aiproject/label_test.py
/aiproject/Testing/bench1.jpg
/aiproject/output_labels.txt
/aiproject/retrained_graph.pb
Once the model is trained, the retrained_graph.pb file and the labels.txt files should be linked to the android application. The detailed steps and the pertaining files can be found on Shantanu's Github, who was responsible for the android application developement.
The apk file can be accessed here.