Training Detection Networks

To create an image database for training a faster-RCNN detector:

1. Select "Tools > Network Training > Create Network Training Images"

2. In the dialog box, select all files to create images from.

3. Enter spectrogram settings.

   • FTT windows length, overlap, and NFFT are specified in seconds.

     • Tip: Set Focus Window to the image length you want to create and use the auto spectrogram settings in the display settings button to determine the best settings.

4. Enter image length.

   • Images should fit a couple calls comfortably. 1/2 second good for short USVs

5. Select number of augmented duplicates.

   • The number of times to duplicate each image, with random amplitude and scaling modification.
   • This feature increases the size and variability of training sets to increase generalizability.

To train a Yolo V2 detector: (Use 2020a or later)

1. Select "Tools > Network Training > Train Network"

2. In the dialog box, select all training tables from which to train from (saved in "DeepSqueak\Training\").

3. Decide whether or not to use a pre-trained network as the starting point.

4. Training will take hours. When finished a save dialog will appear.

For advanced users: To change the network architecture, edit "TrainSqueakDetector.m"