Kumar Desappan

Papers

Total Citations

H-Index

About

Kumar Desappan is a leading researcher in embedded computer vision and deep learning acceleration, with a focus on enabling efficient AI inference on resource-constrained heterogeneous systems. His work addresses the critical challenge of deploying complex neural networks on devices ranging from automotive platforms to industrial machine vision and robotics. Desappan’s major contributions include pioneering dynamic and predictive quantization methods for CNN inference, a technique that minimizes accuracy loss while dramatically reducing memory and computational demands for embedded deployment. This work, cited 6 times, is foundational for real-time, on-device AI. He also developed a novel OpenVX implementation for heterogeneous multi-core architectures (CPU, GPU, DSP, HWA), achieving high utilization and low latency across diverse compute elements—a key enabler for modern vision pipelines in autonomous systems and AR/VR. With 5 citations, this framework demonstrates his impact on standardizing cross-platform computer vision. Desappan’s research directly bridges the gap between cutting-edge deep learning algorithms and practical, real-world embedded systems, making him a pivotal figure in the democratization of efficient, on-device intelligence.