Lidar sensors excel in providing high precision measurements of the three-dimensional environment. These sensors are a key component of any autonomous vehicle today. However, the large amount of data that they provide makes it intractable for real-time applications, such as autonomous driving. This has pushed many authors in research and industry to use compressed representations for the lidar data, such as voxelization or bird's-eye-view, which makes computations real-time at the cost of introducing quantization errors and losing information. This project explores the image view representation of lidar data as an alternative to the aforementioned representations, as this representation does not remove any information, and it is two-dimensional, which shall allow for real-time processing of the data while keeping all the original information. Although there have been some attempts in literature to use the lidar image view for object detection, these have mostly consisted of convolutional neural networks, and none has used a transformer architecture for it, which is potentially more powerful as has been proved in other applications. The goal of this project is to investigate the feasibility and performance of transformer architectures on the lidar image view representation for object detection tasks. These will be benchmarked against public datasets and compared to other object detection methods that operate on different data representations and have different architectures.