Esteban Pesnel’s homepage
I'm a PhD student jointly affiliated with INRIA (team CompACT) and MediaKind in Rennes, France. My research sits at the intersection of deep learning and video compression - designing neural networks that make standard codecs more efficient.
Before starting my PhD, I graduated from INSA Rennes in electronics & computer engineering, with an exchange semester in aerospace engineering at ÉTS Montréal.
My research focuses on end-to-end rate-distortion optimization of video streaming systems by learning a neural pre-processing \(f\) and optionally a post-processing \(g\) (which can be learned or fixed, e.g. bicubic) around a standard video codec \(\phi\):
The central challenge is that conventional codecs (H.264, HEVC, ...) are non-differentiable - gradients cannot flow through them, preventing standard backpropagation across the full pipeline. My work explores strategies to overcome this barrier: surrogate gradients, differentiable proxies, and codec-aware training schemes.
📄 HAL preprint · 🔍 Details
🔗 Try it · 📦 GitHub · 📋 Projects