Glimpse of MCQ Based VQA in Road & Traffic Scenarios

Multi-modal models have been a boon to the commu-nity over the last decade. The Large Language and Vision Models for Autonomous Driving (LLVM-AD) challenge [3] is a Visual Question Answering (VQA) problem in Multiple Choice Question (MCQ) answering in road/traffic scenarios. We have attempted this as a supervised clas-sification task. In this exercise, we have explored vari-ous models, from transformer-based multi-modal models to recent Vision Language Models (VLMs). Even though we have experimented with time and compute constraints, the results show that the models could enhance the ca-pability if trained for longer iterations. We have also compared the model sizes considering the limited tar-get edge compute. Codes are available on the following https://github.com/Athirakr94/VQA.git.