Bishal Swain

Bishal Swain

Deep Learning Researcher in the Making.

Publications

You can also find my articles on my Google Scholar profile.
Upcoming Conferences

Conference Papers

Coupling Liquid Time-Constant Encoders with Modern Hopfield Memory

Published in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026

Continuous-time neural networks often struggle to preserve long-term context because a single hidden state must encode both fast-changing inputs and slower temporal information. We address this by coupling a Liquid Time-Constant Network (LTC) with a Modern Hopfield Network (MHN) for associative memory retrieval. The resulting LTC-MHN model produces bounded representations, smooths optimization through gradient contraction, and improves performance on time-series benchmarks.

Download Paper

Guided Texture Segmentation via Coordinate-Aware Class-Ratio Mapping

Published in IEEE/CVF Winter Conference on Applications of Computer Vision [WACV], 2026

We introduce a guided segmentation framework for texture-rich images that leverages coordinate-aware class-ratio mapping to incorporate global distributional priors into pixel-level predictions. Expected class proportions are transformed into spatial maps and fused with encoder representations through an adaptive gate, enforcing consistency between global composition and local evidence. This conditioning enables the model to resolve ambiguous textures commonly found in metallographic SEM images.

Download Paper

Out-of-Distribution Nuclei Segmentation in Histology Imaging via Liquid Neural Networks with Modern Hopfield Layer

Published in Medical Image Computing and Computer Assisted Intervention [MICCAI], 2025

We present a nuclei segmentation framework that integrates Liquid Neural Networks with a Modern Hopfield Layer to improve robustness under distribution shifts in histology imaging. By processing embeddings in reverse hierarchical order and stabilizing them through associative memory, our method enhances domain-invariant representations. Experiments on benchmark datasets show an average 16.35% OOD improvement over baselines.

Download Paper

SAM Guided Task-Specific Enhanced Nuclei Segmentation in Digital Pathology

Published in Medical Image Computing and Computer Assisted Intervention [MICCAI], 2024

In this paper we propose utilizing the foundation model to guide the task-specific supervised learning by dynamically combining their global and local latent representations, via our proposed X-Gated Fusion Block, which uses Gated squeeze and excitation block followed by Cross-attention to dynamically fuse latent representations.

Download Paper

Journal Articles