Consists of two tasks: stitching images and using CNN for planes and birds recognition.
See here.
For CNN used LeNet architecture.
Usage example:
Go to planes_birds directory:
cd planes_birds
Train and save model:
python lenet_planes_birds.py --save-model 1 --weights pb_rmsp.hdf5
Load already saved model:
python lenet_planes_birds.py --load-model 1 --weights pb_rmsp.hdf5
Use SGD optimizer (default one is RMSProp):
python lenet_planes_birds.py --save-model 1 --weights pb_sgd.hdf5 --sgd-optimizer 1
Use Adadelta optimizer (default one is RMSProp):
python lenet_planes_birds.py --save-model 1 --weights pb_adadelta.hdf5 --adadelta-optimizer 1
Train on gray images:
python lenet_planes_birds.py --save-model 1 --weights pb_sgd.hdf5 --gray-scale 1
Results:
| Accuracy | RMSProp | SGD | Adadelta |
|---|---|---|---|
| Color image | 70.50% | 77.50% | 76.30% |
| Gray image | 56.90% | 62.60% | 57.00% |