Binary Vehicle Classifier

The EDA, model development, and evaluation of a binary image classifier to detect emergency vehicles.

Getting started

There are several notebooks that contain starter code for the model and also explains the dataset version that was used to train a binary emergency vehicle classifier. The dataset and weights reside on the EC2 instance called Training_Big_Boy. To ssh into this instance, you will need the justin_training_gpu.pem file, which can be found in an S3 bucket called lokoai-it-security.

There are several notebooks in this repo, but the two most important are listed below:

• emergency_vehicle_eda.ipynb contains EDA related to the emergency_vehicle dataset
• emergency_vehicle_model_dev.ipynb has starter code to building the model and loading data
• yolo_plus_emergency_dev.ipynb a PoC to see the results of linking a pre-trained yolo model to the emergency vehicle classifier. Results weren't satisfactory.

Next Steps

A simple way to improve the models performance is to switch to a much more powerful SSD model, such as CLIP. Example code for using CLIP as a single shot detector is located in the CLIP_dev.ipynb