The model described in this card is a classification network which aims to classify car images into 6 vehicle types:
This is a classification model with a Resnet18 backbone.
The training algorithm optimizes the network to minimize the categorical cross entropy loss for the classes. The training is carried out in two phases. This model was trained using the Image Classification training app in TAO Toolkit v3.0. In the first phase, the network is trained with regularization to facilitate pruning. Following the first phase, we prune the network removing channels whose kernel norms are below the pruning threshold. In the second phase the pruned network is retrained.
VehicleTypeNet is generally cascaded with DashCamNet or TrafficCamNet for smart city applications. For example, DashCamNet or TrafficCamNet acts as a primary detector, detecting the objects of interest and for each detected car the VehicleTypeNet acts as a secondary classifier determining the type of the car.
The datasheet for the model is captured in it’s model card hosted at NGC.