Kitchen utensils detection

This jupyter notebook shows how to preprocess, augment and classify kitchen utensils (bowl,plate) images using inception model

Dataset

Images of these utensils are manually collected and placed under 2 categories i.e. bowl and plate. A third class known as negative contain images of no-interest that do not include any utensil. A few sample images are shown below:

Bowl

Plate

Negative

Preprocessing

Following preprocessing techniques are used before training and inference on the test data:

1. Images are resized into fix shape
2. Gaussian filter is applied on resized image to the point where our object is still visible through naked eye

Model training

Pre-trained Inception Network is used to extract features from video frames. A fully-connected layer based on number of classes i.e. 3 is used at the end and network is trained on augmented data