---

1. To create a dog classification model, we re-trained resnet50 through transfer learning We thought Resnet was a well-learned model, so we thought it was possible just to re-learn a fully connected layer

2. The following shows the code that Google Corab used to transfer the resnet50 and the resultant model saved as a pt file. We categorized the dogs with the saved model files

3. Five types of fierce dogs are combined to divide classes into fierce dogs and non-fierce dogs to check their accuracy

4. We learned the model in the way we chose, and as a result, we found that they were good at judging whether or not they were fierce dogs, as shown in the pictures