An image classification and detection project on dog and cat images. From the given single-object image, we need to classify the image into 2 classes:

• Dog class
• Cat class

Furthermore, we also try to detect the position of that object on the image (localization).

We use the dataset from Kaggle: Dogs vs cats

In this project we try four models of VGG-style, including:

• VGG 1-block
• VGG 2-block
• VGG 3-block Each block of the molde contains a pair of Convolute2D and Pooling, then followed by an Fully-connected layer.

Moreover, we also leverage the VGG-16 model to do transfer learning. We use the pretrained VGG-16 model without the top layer and train the new output layer (fixed weights for other layers).

To do the detection task, we use the very idea which is the sliding window. The algorithm try different of windows and slide on the image to find the most-condfident zone.

This project was developed by our team, including:

• phuc16102001
• thanhoang4869
• buingocchinh321

MIT