#4 Working on Tensorflow Image Classification with Transfer Learning - Improving Model

#machinelearning #machinelearningproject #kaggle #tensorflow

How often do you have a hawk landing on your roof? Not very often, right? That's why I need to get my wildlife camera more attuned so I don't miss out.

Yesterday in my Livestream I finally managed to gather a test data set with images captured by my wildlife camera. I then did an evaluation of the model that I trained with images that I managed to gather from other datasets on Kaggle and from other websites on the internet.
I decided not to use images from my camera for training because I wanted to prevent my machine learning model from overfitting to the background in the photos, which happens to be my garden.

When training a machine learning model there are three types of datasets that you typically use.

The training data set contains images that are provided to the model during training. The model learns from these images.
The validation dataset is not used by the model to learn but it's used during training, to score how well the neural network is learning. This helps in the training process by guiding in which direction to change the hyperparameters and weights in the model.

The model I trained using a dataset that didn't include images from my raspberry pi camera didn't do too well. So I need to include some images from my Pi in the training process. I will first try to use PI câmeras for the validation dataset., and see if that is enough to improve my results.