Making logs more interpretable.

- Solving stroke using AI.- Finding biomarkers for neurodegenerative diseases using AI.

Advisors: Prof. Mohamed H. Elhoseiny and Prof. Peter Wonka• Developed 3D CoMPaT, a richly annotated large‐scale dataset of more than 7.19 million rendered compositions of Materials on Parts of 7262 unique 3D Models; 990 compositions per model on average. Published this work in ECCV 2022 Oral Track.• The rendering involved using blender and testing various materials and selecting a few good materials for the available object set. 193 materials were selected belonging to 13 broad material… Show more Advisors: Prof. Mohamed H. Elhoseiny and Prof. Peter Wonka• Developed 3D CoMPaT, a richly annotated large‐scale dataset of more than 7.19 million rendered compositions of Materials on Parts of 7262 unique 3D Models; 990 compositions per model on average. Published this work in ECCV 2022 Oral Track.• The rendering involved using blender and testing various materials and selecting a few good materials for the available object set. 193 materials were selected belonging to 13 broad material super‐categories. The results were checked in a more powerful renderer ‐ 3dsmax.• Proposed a novel problem for detecting and segmenting material over parts of a 3D object. Developed baseline using several state-of-the-art methods like PointGroup, BPNet for segmenting parts and materials.• Developed a novel transformer-based method in visual scene understanding in 3D. Introduced multi‐modal transformer approach to develop a shared latent space between 3D scene and language description.• We achieved state of the art results on Referit3D, Nr3D, Sr3D and published the results in WACV 2022. Show less

Advisors: Dr. Anupama Ray (IBM Research), Dr. Ritu Garg (Intel AI), Dr. P. Mukherjee (JNU, IIT Delhi)- Develop Fine-grained Aggressive Behaviour detection in videos using Deep Neural Networks on UCF-Crimes Dataset using use self-attention models with transformers and dynamic frame sampling neural network. Performed various ablation studies on the UCF-Crimes dataset.- Our Model ended up beating state-of-the-art on the dataset by 1% with an accuracy of 29.2%.

Advisors: Dr. Aakanksha Chowdhery (DeepMind) And Dr. Prerana Mukherjee (IIT Delhi, JNU).- The project was developed under The Marconi Society and funded by Google.- Developed a neural network for voice stress detection. This work focused on providing a model which is capable of working on edge devices.• Both keyword spotting and emotion recognition models got an accuracy of 92%. A speech dataset called EmoSpeech Command Dataset was also curated. Dataset Link:… Show more Advisors: Dr. Aakanksha Chowdhery (DeepMind) And Dr. Prerana Mukherjee (IIT Delhi, JNU).- The project was developed under The Marconi Society and funded by Google.- Developed a neural network for voice stress detection. This work focused on providing a model which is capable of working on edge devices.• Both keyword spotting and emotion recognition models got an accuracy of 92%. A speech dataset called EmoSpeech Command Dataset was also curated. Dataset Link: https://emo-speech.web.app/. The project has received national media coverage for aiding socio-economic causes. Show less

AI Research Grant received:- Early Innovation Research Grant- OpenVino Early Innovation Research GrantBoth grant were in form of $5000 + hardware support.