############SPACEWHALE Project: Whales detection based on deep learning method###########
#########################################################################################
The object detection is done using the “Faster R-CNN" (Region-based Convolutional Neural Network) algorithm.

#######Cross validation:
To test the algorithm on independent data, a four-fold cross validation technique was used where the four satellite images were each considered a fold. To do this, the algorithm was trained using whales from 3 of the satellite images plus the downsampled minke whales from the digital aerial footage. The algorithm was run on the held back 4th image, and evaluation statistics were calculated from this. This was run 4 times, holding back one larger image in each iteration.

we have 4 large satellite images annotated as 1, 2, 3 and5. Usually we train with 3 large satellite image and test in the 4th one so for example training with images 1, 2, 5 and testing with image 3:
##### training step:
python3 train.py --model_path './model/' --num_classes 5 --batch_size 18 --num_epochs 30

##### testing atep:
python3 test.py --model_path './model/resnet50_img125.pth' --input_path 'image3' --output_path './outputs/image3' --box_score 0.01