featured

Generative Multimodal Learning for Reconstructing Missing Modality

Training a latent variable based variational inference model on multimodal data in order to perform inference with all possible combinations of missing modalities.

Policy Gradient

Reproducibility and Analysis of Deep Policy Gradient methods for Reinforcement Learning Tasks

Online Learning of temporal Knowledge Graphs

In this project, we apply currently available solutions to address incremental knowledge graph embedding to several applications to test their efficiency.

In this project, we propose an incremental learning problem for Knowledge Graphs to obtain representations for new entities and also update the representations of old entities that share facts with these newer entities.

Image Stitching (Panorama)

Implemented an image stitching algorithm for creating panoramas from successive images from a rotating camera from scratch.

Generic Extraction Module (G.E.M)

Trained a biLSTM model using both word and character level embeddings for information retrieval from text OCR outputs of ID cards

Image Quality Assessment

An ensemble model to quantify image quality to filter poor quality images at the client end to prevent redundant processing

Dory OCR

We created a state of the art Optical Character Recognition Engine specifically for Indian ID cards using a pipeline for document layout detection, foreground extraction, text detection, recognition and postprocessing

Sign Language Classification [Bachelor Project]

Deep Learning based Indian Sign Language detection for conversion to speech, a subsystem of the Hindi speech-Indian sign language interconversion system

Activity Recognition

Using traditional computer vision with deep learning algorithms for Anomalous activity detection from CCTV camera feed