Climate Change AI Hackathon

How to run?

• Install python deps using pip install -r requirements.txt.
• Download train.zip and test.zip from bit.ly/2mgYsqh, extract into a folder named data.
• Run create_tfrecords.py to create training tfrecord. Run slim/train_pc.sh to train model. The loss profile should look something like below.

• Run slim/run_on_test.sh to generate predictions on test set.
• slim/eval_pc.sh can be used to check accuracy on training set or validation set.
• The flask app for monitoring is run by python3 app/app.py.
• slim/export_frozen_graph.sh and slim/export_inference_graph.sh to be used for converting saved TF models to frozen graph and tflite_converter.py to convert to TFLite.
• slim/train_logs has the exported frozen graph as well the JSON file with predictions on the test set.
• Android app code inside android.

Flask app for visualizing GPS coords of images taken and their predictions

3D sparse map from VO (outdoor)

3D sparse map from VO (indoor)

Live demo during hackathon