Dear Rahul Jain ,

Vision Transformers (ViTs), with the magnificent potential to unravel the information contained within images, have evolved as one of the most contemporary and dominant architectures that are being used in the field of computer vision.

https://www.finextra.com/blogposting/26447/vision-transformer-in-computer-vision-transforming-the-way-we-look-at-images

Regards,

Shafagat

Thanks for opening up such an interesting topic for discussion. ViT is a new record-breaking paradigm in computer vision. The domain of computer vision has been widely dominated by CNNs until quite recently, mostly because they model local patterns very well. On the other hand, a ViT applies the self-attention mechanism while capturing long-range dependencies and is hence able to process global information more effectively when compared to CNNs. This mechanism has been seen to perform aptly in image classification, object detection, and segmentation tasks with state-of-the-art performance levels. Furthermore, it is very much adaptable and scalable in handling different modalities like images, videos, and texts: this makes ViTs very important in the current era (Khan et al., 2021). As research progresses, ViTs show dramatically improved performance with large-scale datasets—a fact strongly evidenced by the models' growth to billions of parameters and their achievements in state-of-the-art tasks, such as ImageNet classification (Zhai et al., 2021). All these characteristics make ViTs a versatile tool for modern AI, exhibiting flexibility in low-level vision applications, such as superresolution or image enhancement, and flexibility in multimodal settings, such as visual question answering. Expected innovations will probably be in the direction of increasing architectural efficiency while expanding the capabilities to solve ever more complex problems in computer vision.

• Khan, S. H., Naseer, M., Hayat, M., Zamir, S. W., Khan, F. S., & Shah, M. (2021). Transformers in Vision: A Survey. ACM Computing Surveys (CSUR).
• Zhai, X., Kolesnikov, A., Houlsby, N., & Beyer, L. (2021). Scaling Vision Transformers. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).