planes-detector is an ergo based project that relies on a convolutional neural network to detect airplanes from satellite imagery.

Download the PlanesNet dataset (the zip file with pictures) and extract it somewhere, then:

ergo train /path/to/planes-detector --dataset /path/to/planesnet/pictures

This will make ergo preprocess the pictures, import them as a csv dataset and start the training algorithm (see how to train on GPU) that can be monitored with tensorboard (logs folder).

After training is completed, you can view the model structure and how the accuracy and loss metrics changed during training with:

ergo view /path/to/planes-detector

Training	ROC/AUC

Training	Validation	Testing

Once the training is completed, you can clean the project from temporary datasets and start a REST API server to test the model:

cd /path/to/planes-detector
ergo clean .
ergo serve .

You can test the predictions with curl:

curl https://127.0.0.1:8080/?x=/path/to/filename.jpg

Alternatively you can use the file model.h5 (created inside the project folder after training) by loadeding it as you would normally do with Keras API for evaluation.

planes-detector was made with ♥ by Simone Margaritelli and it is released under the GPL 3 license.