USING BILINEAR CNNs FOR VEHICLE MAKE AND MODEL PREDICTION (2022)
- Conducted fine-grained vehicle classification research to predict vehicle make and model from input images using various neural network architectures. The project utilized the VMMRdb dataset as the primary data source for training and evaluation.
- Evaluated three distinct approaches: transfer learning with backbone models (ResNet18, ResNet50, MobileNetv2), Bilinear CNNs, and Vision Transformers. As the number of classification labels increased, Bilinear CNNs demonstrated superior performance in accuracy, effectively capturing fine-grained details that distinguish between vehicle models. The research findings are summarized in the poster below.
- This research was conducted as part of CS 7643 (Deep Learning). The complete implementation is available in the code repository, along with the detailed research report.