Vishal Keshav

Vishal Keshav, Fabien Delattre

Self-supervised learning techniques have shown their abilities to learn meaningful feature representation. This is made possible by training a model on pretext tasks that only requires to find correlations between inputs or parts of inputs. However, such pretext tasks need to be carefully hand selected to avoid low level signals that could make those pretext tasks trivial. Moreover, removing those shortcuts often leads to the loss of some semantically valuable information. We show that it directly impacts the speed of learning of the downstream task. In this paper we took inspiration from curriculum learning to progressively remove low level signals and show that it significantly increase the speed of convergence of the downstream task.

Vishal Keshav, Tej Pratap GVSL

This paper presents a novel design methodology for architecting a light-weight and faster DNN architecture for vision applications. The effectiveness of the architecture is demonstrated on Color-Constancy use case an inherent block in camera and imaging pipelines. Specifically, we present a multi-branch architecture that disassembles the contextual features and color properties from an image, and later combines them to predict a global property (e.g. Global Illumination). We also propose an implicit regularization technique by designing cross-branch regularization block that enables the network to retain high generalization accuracy. With a conservative use of best computational operators, the proposed architecture achieves state-of-the-art accuracy with 30X lesser model parameters and 70X faster inference time for color constancy. It is also shown that the proposed architecture is generic and achieves similar efficiency in other vision applications such as Low-Light photography.