my publications.
Enhanced Compression Model (ECM): A Hyperprior-Based Neural Image Compression Framework
IEEE Xplore, presented at INSTCon 2026, NIT Rourkela
Extended a Scale Hyperprior VAE architecture with semantic gating, channel attention, autoregressive context modelling, an entropy parameter network, and scale refinement. Achieved 0.176 BPP, 30.882 dB PSNR, and 0.985 MS-SSIM on the Kodak benchmark, outperforming the Scale Hyperprior baseline and classical codecs (JPEG, WebP) at comparable or lower bitrates.
Computer Vision
Deep Learning
Neural Compression
VAE