cs231n project

In this project, a new approach that classifies the building instance category by using the street view images is presented. Since every building is unique, the classical image classification architectures have limited performance on this particular task. Therefore, the project proposes a new model that takes account of both building’s design and region. In order to achieve this, a multi-task training technique considering the building's region label is implemented on top of a classical ResNet50 model.

Running the Code

put the dataset in the 'dataset' directory before training

Train and evaluate the model

python main_v2.py

Test the model

python evaluate_v2.py

build the city map

python city_map_builder.py

Authors

• Zonglin Li - Zonglin
• Yanlong Ma - Yanlong