Radhakrishna Achanta

Timeline

1. 2009

   Frequency-tuned salient region detection

   He and three coauthors, Sheila S. Hemami, Francisco J. Estrada and Sabine Süsstrunk, present a method for salient region detection at CVPR in 2009. It outputs full resolution saliency maps with well-defined boundaries of salient objects and keeps more frequency content from the original image than other existing techniques.

2. 2010–2012

   The SLIC superpixel algorithm

   With Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua and Sabine Süsstrunk, he introduces SLIC (Simple Linear Iterative Clustering), which clusters pixels in a five-dimensional color and image plane space to generate compact, nearly uniform superpixels. The work appears as an EPFL technical report in 2010 and in IEEE Transactions on Pattern Analysis and Machine Intelligence in 2012.

3. January 2011

   PhD thesis at EPFL

   He defends his doctoral thesis on finding objects of interest in images using saliency and superpixels at EPFL in January 2011, with Sabine Süsstrunk as thesis director. His degrees are a PhD in computer science from EPFL, an MSc in computer science from NUS Singapore and a BEng in electrical engineering from JEC India.

4. 2017

   Non-iterative superpixels and polygons

   At CVPR in 2017, he and Sabine Süsstrunk present superpixels and polygons using Simple Non-Iterative Clustering, an improved version of SLIC that is non-iterative, enforces connectivity from the start, requires less memory and is faster. The paper also presents a polygonal partitioning algorithm built on the superpixel boundaries.

5. By 2022

   Principal data scientist at SDSC

   He is a principal data scientist at the Swiss Data Science Center, where his projects include TAPEDA on post-earthquake damage assessment and DeepMICROIA on deep learning for TOMCAT imaging. In 2022, he coauthors the NeurIPS paper What You See is What You Classify: Black Box Attributions, on unboxing deep learning models for image classification, with Steven Stalder, Nathanaël Perraudin, Fernando Perez-Cruz and Michele Volpi.

6. 2023–2026

   Diffusion models at EPFL

   In 2023, he coauthors Exploiting the Signal-Leak Bias in Diffusion Models with Martin Nicolas Everaert, Athanasios Fitsios, Marco Bocchio, Sami Arpa and Sabine Süsstrunk, which exploits a signal-leak bias in existing diffusion models to give more control over generated images. At EPFL's Image and Visual Representation Laboratory, he co-directs the doctorates of Nikolaos Arvanitopoulos Darginis (2017), Bryan German Pantoja Rosero (2023) and Martin Nicolas Everaert (2026).

7. Since 2025

   Deep learning tech lead at Cisco

   From 2025, he works as a deep learning tech lead at Cisco Systems. In June 2026, he and Will Reed post TallyTrain, a federated distillation protocol that matches or beats soft-label distillation with up to three orders of magnitude less communication. In July 2026, they follow with FLITE (Federated Low-rank Iterative Training Engine), a federated fine-tuning method that sends 1,280 floats, about 5 KB, per client per round on CIFAR-100.