Prithwish Jana

Jadavpur University, Indian Institute of Technology Kharagpur

Papers

Total Citations

H-Index

About

Prithwish Jana is a researcher at the intersection of computer vision and deep learning, with a focus on semantic segmentation and video understanding. His work addresses fundamental challenges in visual perception, particularly in adapting complex models to real-world, unconstrained data. In his 2020 paper on semantic segmentation, Jana proposed a two-stage pipeline that combines the statistical efficiency of SEEDS superpixels with the representational power of a ForkNet deep architecture. This hybrid approach aimed to bridge the gap between low-resource statistical methods and high-accuracy deep learning, offering a practical solution for labeling diverse natural environments where purely statistical methods fail. His 2021 research tackles a critical bottleneck in video analysis: the preprocessing of untrimmed, variable-aspect-ratio videos for 3D ConvNets. By introducing an unsupervised action localization crop for video retargeting, Jana’s work ensures that the primary subject of a video is preserved during the mandatory square-cropping step, preventing information loss common in random or center-cropping strategies. While his citation counts (4 and 3, respectively) reflect early-stage impact, these contributions demonstrate a clear trajectory toward making deep learning models more robust and practical for real-world deployment in surveillance, robotics, and social media analysis.