"Car Orientation Convolutional Network"
A convolutional neural network aimed at determining the compass direction of cars.
Use cases are to detect cars and to determine driving direction on roads.
Access dataset through: EPFL
Labels were not provided, and so they were calculated by looking at the picture taking rate versus the rotation rate.
- Image sizes are (376,250)
- 20 image sequences
- Labels are sometimes within 1-2° of true label since times were truncated to whole seconds.
- Transfer Learning on AlexNet
- Histogram of Oriented Gradients Feedforward Network
Both networks learn car pose angles via a discretization and probability distribution approach according to Hara, Vemulapalli, and Chellappa (2017).