Research

Consider a questions like "What is written on the signboard?", the answer predicted by the model is always" STOP" which makes the model to ignore the image. To address these issues, we propose a method to learn visual features (making V matter in TextVQA) along with the OCR features and question features using VQA dataset as external knowledge for Text-based VQA. Specifically, we combine the TextVQA dataset and VQA dataset and train the model on this combined dataset. Such a simple, yet effective approach increases the understanding and correlation between the image features and text present in the image, which helps in the better answering of questions. 

Growing interest in conversational agents promote two-way human-computer communications involving asking and answering visual questions have become an active area of research in AI. Thus, generation of visual question-answer pair (s) becomes an important and challenging task. To address this issue, we propose a weakly-supervised visual question answer generation method that generates a relevant question-answer pairs for a given input image and associated caption. Most of the prior works are supervised and depend on the annotated question-answer datasets. In our work, we present a weakly supervised method that synthetically generates question-answer pairs procedurally from visual information and captions.

TextVQG aims to generate a natural language question for a given input image and an automatically extracted text also known as OCR token from it such that the OCR token is an answer to the generated question.

A study to collect sound samples for developing non-invasive diagnosis of respiratory diseases and COVID-19.

The goal of this track is to evaluate the performance of different retrieval algorithms under the Open-Set setting and modality-missing setting, respectively. Since objects from unseen categories are very common in real-world applications, we design the open-set 3D object retrieval to expand the application of traditional 3D object retrieval.

Unlike human vision, supervising the machine to analyze segments of an object is challenging & quite essential in various machine vision applications. We propose PIG-Net for 3D part segmentation of point clouds based on inception.

Scene understanding of large-scale 3D point clouds of an outer space is a challenging task. We propose GRanD-Net for  semantic segmentation of large-scale 3D point clouds of street scenes.

3D semantic segmentation is one of the most challenging events in robotic vision tasks for detection and identification of various objects in a scene. We propose a lightweight semantic segmentation network with dilated convolutions.

As powerful computation resources and scanning devices have led to an exponential growth of 3D point cloud data, retrieving the relevant 3D objects from databases is a challenging task. We propose the necessity of 3D feature encoding along with local descriptors for solving non-rigid and rigid point cloud retrieval.

We propose a 3D object categorization framework comprising of IWKS & MTCS feature descriptors by approximating Laplace-Beltrami operator on point cloud data for non-rigid objects and MTCS for rigid objects.

In recent years, different methods have relied on the underwater image formation model and deep learning techniques to restore the underwater image, but tend to produce unnatural artifacts and reduced levels of sharpness. To address these challenges, we proposed FloodNet using residual dense learning with the objective of estimating restored underwater images from a wide variety of degraded underwater images.

We propose end-to-end deep convolutional network architecture to restore the underwater degraded images and improve visual perception by utilizing the symmetric skip connections between encoder–decoder. 

With the advent of modern digital technology there is a great surge of interest among the computer graphics and vision community in digital restoration of 3D models and many problems still exist in building the complete 3D reconstruction framework. In our research work, we address the problems of digital restoration of 3D models at cultural heritage sites specifically artifacts at Hampi, India. We propose to build a 3D reconstruction pipeline, referring to acquisition, data processing and rendering for digital restoration of 3D models at cultural heritage sites.

We address the problem of 3D inpainting using ROI-based method for point clouds. We focus on inpainting of complex, irregular and large missing regions covering prominent geometric features by considering n self-similar examples.