Publications
Publications
Selected papers across reliable adaptive vision, continual learning, and deployment-aware perception.
For recent accepted papers, direct venue and preprint links are provided. Additional artifacts are added as they become public.
Publications
AdaPrior: Bayesian-Inspired Adaptive Prior Correction for Long-Tailed Continual Learning
Authors: S Divakar Bhat, Amit Popat More, Mudit Soni, Bhuvan Aggarwal
Venue: IEEE/CVF CVPR 2026 Main Conference (Highlight)
A method for long-tailed continual learning that adaptively corrects model priors to reduce bias under class imbalance and incremental distribution shifts.
Consistent Yet Wrong: Evidence Insensitivity in Spatial Vision-Language Models
Authors: S Divakar Bhat, Toshihiko Yamasaki
Venue: Workshop on Multimodal Learning and Applications (MULA), IEEE/CVF CVPR Workshops 2026
An analysis of evidence-insensitive behavior in spatial VLMs, highlighting consistent predictions that remain poorly grounded to visual evidence.
Seeing What’s Not There: Negation Understanding Needs More Than Training
Authors: B. Aggarwal, A. More, M. Soni, S. D. Bhat
Venue: ICLR 2026
Shows limits of current VLM training regimes for negation understanding, motivating stronger reasoning-oriented adaptation and evaluation protocols.
Official paper/code links will be added once publicly available.
Show more publications
Prior2Posterior: Model Prior Correction for Long-Tailed Learning
Authors: S. D. Bhat, A. More, M. Soni, S. Agrawal
Venue: IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025 (Oral)
Introduces model prior correction to improve recognition reliability in long-tailed settings.
PC-GZSL: Prior Correction for Generalized Zero Shot Learning
Authors: S. D. Bhat, A. More, M. Soni, B. Aggarwal
Venue: IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025
A prior correction method for generalized zero-shot learning under class imbalance and calibration mismatch.
Model Ensemble to Fuse Geometric and Learning Solutions for Camera Rotation Estimation
Authors: B. Aggarwal, A. More, S. D. Bhat, M. Soni
Venue: IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2025
Combines geometric and learning-based components for robust camera rotation estimation.
Robust Loss Function for Class Imbalanced Semantic Segmentation and Image Classification
Authors: S. D. Bhat, A. More, M. Soni, Y. Yasui
Venue: IFAC World Congress, 2023
Presents a robust objective for class-imbalanced segmentation and classification problems.
CILEA-NET: Curriculum-Driven Incremental Learning Network for Remote Sensing Image Classification
Authors: S. D. Bhat, B. Banerjee, S. Chaudhuri, A. Bhattacharya
Venue: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021
Introduces a curriculum-based incremental learning framework for robust remote sensing scene classification.
SemGIF: A Semantics Guided Incremental Few-shot Learning Framework with Generative Replay
Authors: S. D. Bhat, B. Banerjee, S. Chaudhuri
Venue: BMVC 2021
Combines semantic guidance with generative replay for incremental few-shot learning under evolving class distributions.
Directed Variational Cross-encoder Network for Few-shot Multi-image Co-segmentation
Authors: S. Banerjee, S. D. Bhat, S. Chaudhuri, R. Velmurugan
Venue: 25th International Conference on Pattern Recognition (ICPR), 2021
A variational cross-encoder formulation for data-efficient co-segmentation in few-shot multi-image settings.