Valay Bundele

Publications
1. V. Bundele*, D. Pal*, B. Banerjee, Y. Jeppu; ‘SPN: Stable Prototypical Network for Few-Shot
Learning-Based Hyperspectral Image Classification’, IEEE Geoscience and Remote Sensing Letters
2. D. Pal, V. Bundele, R. Sharma, B. Banerjee, Y. Jeppu; ‘Few-Shot Open-Set Recognition of Hyperspectral
Images with Outlier Calibration Network’, IEEE Winter Conference on Applications of Computer Vision
(WACV), 2022
3. S. Nasser, S. Shamsi, V. Bundele, B. Garg, A. Sethi; ‘Perceptual cGAN for MRI Super-resolution’, IEEE
Engineering in Medicine and Biology Conference 2022
4. M. Singha, B. Solanki, A. Jha, V. Bundele, B. Banerjee, S. Chaudhari; ‘AppleNet : Attention
Parameterized Prompt Learning based Enhanced Network for Few-Shot Image Classification in Remote
Sensing’ (to be submitted to IEEE Transactions on Geoscience and Remote Sensing)

VALAY BUNDELE

Research Interests
Computer Vision, Deep Learning, Learning under Limited Supervision
Education
Indian Institute of Technology Bombay Mumbai, India
B.Tech in Energy Engineering with M.Tech in AI and Data Science July 2018 – June 2023
• Cumulative GPA: 8.77/10.0
• Minor Degree in Computer Science and Engineering
Aklank Public School(CBSE) Kota, India
Intermediate/+2 Percentage:91.40%
Publications
1. V. Bundele*, D. Pal*, B. Banerjee, Y. Jeppu; ‘SPN: Stable Prototypical Network for Few-Shot
Learning-Based Hyperspectral Image Classification’, IEEE Geoscience and Remote Sensing Letters
2. D. Pal, V. Bundele, R. Sharma, B. Banerjee, Y. Jeppu; ‘Few-Shot Open-Set Recognition of Hyperspectral
Images with Outlier Calibration Network’, IEEE Winter Conference on Applications of Computer Vision
(WACV), 2022
3. S. Nasser, S. Shamsi, V. Bundele, B. Garg, A. Sethi; ‘Perceptual cGAN for MRI Super-resolution’, IEEE
Engineering in Medicine and Biology Conference 2022
4. M. Singha, B. Solanki, A. Jha, V. Bundele, B. Banerjee, S. Chaudhari; ‘AppleNet : Attention
Parameterized Prompt Learning based Enhanced Network for Few-Shot Image Classification in Remote
Sensing’ (to be submitted to IEEE Transactions on Geoscience and Remote Sensing)
Research and Professional Experience
Vision-and-Language Navigation | Dual Degree Project May 2022 - Present
Guide: Prof. Biplab Banerjee and Prof. Aditya Grover (University of California at Los Angeles) IIT Bombay
• Developing agents which can navigate an environment using natural language instructions and visual inputs
• Conditioned Decision Transformer on text instruction, past states and actions for sequential action prediction
• Implemented various text and image encoding schemes involving the use of cross-attention modules to learn effective
joint representations in the BabyAI environment and achieved success rate > 0.99 on 8 levels in multi-task setup
• Explored Matterport3D Simulator and developing algorithms to improve performance in offline RL setup on R2R
Hyperspectral Image Few-Shot Classification May 2020 - Dec 2020
Guide: Dr. Biplab Banerjee IIT Bombay
• Implemented metric-learning based approaches, namely, Relation Networks, Prototypical Networks and
Matching Networks to learn an embedding space using meta-learning for few-shot HSI classification
• Implemented Model-Agnostic Meta Learning to meta-learn initial weights which can be updated using few samples
• Improved the performance by 4% on Salinas by introducing notion of Monte-Carlo Dropblock for network
regularisation by adding DropBlock between the convolutional layers and optimizing a novel variance loss
Few-shot Open-set Recognition of Hyperspectral Images Jan 2021 - Aug 2021
Guide: Dr. Biplab Banerjee IIT Bombay
• Implemented few-shot open-set recognition techniques like PEELER, SnaTCHer for outlier detection in HSI
• Introduced Outlier Calibration Network to overcome requirement of manually setting rejection threshold
• Proposed novel 3D attention module CBAM3D to highlight important regions in spectral-spatial feature maps
• Developed a novel approach involving VAEs to combat data scarcity issue improving accuracy by 2% on Salinas
Few-Shot Image Classification in Remote Sensing Jan - April 2021
Guide: Dr. Biplab Banerjee and Dr. Subhasis Chaudhari IIT Bombay
• Enhanced prompt-learning based vision-language models to improve generalization performance in few-shot setting
• Conducted extensive experiments to evaluate cross-dataset and domain generalization performance of the model
Image Restoration Jan - April 2021
Guide: Dr. Amit Sethi IIT Bombay
• Implemented 3D-VGG16 network for learning the perceptual similarity between the generated and ground-truth
super-resolved MRI images in the Enhanced Super-Resolution GAN setup
• Denoised ultrasound images by implementing paper titled "Residual Dense Networks for Image Restoration"
• Incorporated Convolutional Block Attention Module in the network for efficient flow of refined features
Software Development Intern | Amazon Web Services May – July 2021
Interned in the Aurora CP team which works on AWS Relational Database Management System
• Worked on graph-based integrated testing framework of Relational Database Service to detect destructive paths
• Utilized Java-callgraph to generate a static call graph and thereby get the function calls made in each function
• Implemented a depth-first search based approach to recursively check for annotation on each function using
Reflection API and classified paths having function calls with a specific annotation as destructive
Key Technical Projects
Age and Gender Estimation in Low-Resolution Surveillance Videos | Inter IIT Tech Meet 10.0
Led a team of 6 members, ranked 5th out of 20+ IITs Mar 2022
• Developed a pipeline that can detect the age and gender of a person from a low-quality surveillance video
• Used Facenet embeddings to obtain 93% accuracy on gender detection and an error of ± 3 years on age detection
• Employed YOLOv5 to detect humans in videos and DeepSort algorithm for human tracking and reidentification
• Extracted faces using RetinaFace and super-resolved them using GFPGAN which uses a generative facial prior
Colorization Transformer | Deep Learning Sept - Nov 2021
• Modified implementation of paper titled “Colorization Transformer" by replacing area interpolation with
learnable transpose convolutional layers in the spatial upsampler for colorization of images using transformers
• Utilized axial self-attention blocks to reduce the computational costs involved in capturing global receptive field
Image-to-Image Translation | CS663: Digital Image Processing Spring 2021
• Implemented Pix2Pix GAN, a variant of conditional GANs, for translating images from aerial to map domain
• Developed a generator with skip connections and used a PatchGAN based discriminator for effective translation
Course Assignments | CS747: Foundations of Intelligent and Learning Agents Aug - Nov 2021
• Simulated a multi-armed bandit and implemented sampling algorithms, epsilon-greedy, UCB, and KL-UCB
• Implemented Howard’s Policy Iteration, Value Iteration and Linear Programming to find optimal MDP policy
Facial Attribute Manipulation using GANs | Institute Technical Summer Project May - June 2020
• Implemented AttGAN using Wasserstein loss to generate visually realistic facial images with desired attributes
• Implemented encoder-decoder model to decode latent representations conditioned on expected features
• Applied attribute classification constraint and reconstruction learning for the correct change of facial attributes
Technical Skills
Programming Languages: C++, Python
Libraries/Frameworks: PyTorch, TensorFlow, Keras, OpenCV, Numpy, Pandas
Software: Microsoft Office, LATEX, Git, VS Code
Coursework
Computer Science: Operating Systems, Logic in CS, Data Structures & Algorithms, Design & Analysis of Algorithms
Mathematics: Calculus, Linear Algebra, Differential Equations, Numerical Analysis
Machine Learning: Introduction to Machine Learning, Advanced ML, Medical Image Computing, Foundations of
Intelligent and Learning Agents, Learning with Graphs, Deep Learning : Theory and Practice
Select Mentoring and Leadership Roles
Software Subsystem Head | Team Rakshak Apr 2020 – Mar 2021
Head of the software subsystem of Team Rakshak which works on design and development of UAVs
• Led and managed a team of 10 members to develop algorithms for the tasks of Object Detection, Object
Recognition, and Mapping of surveyed areas for AUVSI SUAS competition held in Maryland, USA
• Employed YOLOv5 for detection and trained classifiers having ResNet backbone to identify object characteristics
• Implemented PCA to find the orientation of objects and performed image stitching using SIFT to generate a map
Coordinator at Mood Indigo, IIT Bombay | Competitions and LYP Aug - Dec 2019
Asia’s largest college cultural festival; 146,000+ foot-fall with 240+ events
• Member of the team that ideated and revamped the governing rules of 5+ competitions in various genres
• Assisted in conceptualizing and organizing Multicity Competitions in 5+ cities, thus increased outreach
• Invited renowned artists across various cultural genres as judges for the competitions on a strict zero budget
Teaching Assistantships | IIT Bombay
Facilitating smooth course organization, grading papers, mentoring students, conducting tutorials and help sessions
• GNR 638 - Machine Learning for Remote Sensing II, Prof. Biplab Banerjee Aug - Nov 2021
• GNR 638 - Machine Learning for Remote Sensing II, Prof. Biplab Banerjee Aug - Nov 2022
Extra Curricular Activities and Other Achievements
Achievements
• Secured 98.92 percentile in JEE Advanced examination among over 0.2 million candidates (2018)
• Among Top 1% students(nationally) in first stage of IJSO among 46,688 candidates (2014-15)
• Secured All India Rank 18 in STaRT examination conducted by Resonance Eduventures (2015)
• Obtained International Rank 193 in National Science Olympiad conducted by SOF (2013)
Mentorship
• Mentored 2 students at Summer Of Science 2020 to understand concepts of Deep Learning
• Guided 8 students on the project titled “Image Colorization" - Seasons of Code 2021
Others
• Reviewer at IEEE Geoscience and Remote Sensing Letters and Springer Cluster Computing Journal
• Assisted in successful execution of the speaker sessions at the E-Summit conducted by E-Cell (2020)
• Completed a biathlon that consisted of 5 kms running and 7 kms cycling (2019)
• Successfully completed a year-long guitar training under National Sports Organization (2018-19)
• Participated in Dainik Bhaskar’s National Level Newspaper Making Competition (2014)