Adrien Bak Homepage


Adrien BAK



DevTech Engineer Computer Vision / Machine Learning - May 2016 to present day

I am a member of the Nvidia team that develops DriveWorks, a set of libraries that takes advantage of specific hardware targeted at autonomous vehicles. I am both implementing a set of tools and computer vision or Deep Learning methods that take advantage of this hardware, to create robust and high performance applications. At the same time, I am helping Nvidia customers to implement their systems and conducting exploratory research to design the systems of tomorrow.

Keywords : Computer Vision - Deep Learning - Autonomous Vehicles - Embedded Computing - GPGPU


Software Engineer - March 2015 to May 2016

Indeed is the #1 job search engine , serving more than 180 millions of jobseekers every month. As a member of the aggregation team, I was responsible for finding them jobs to apply to. More specifically, I designed and implemented tools based on Machine Learning and Data Mining that crawl and scrape the web to identify job postings in arbitrary websites. Additionally, those tools are responsible for extracting a number of unstructured metadata to make indexing possible. Those tools have to perform reliably, and at scale. In addition to my engineering responsibilities, I also took part in mentoring new hires, in order to help them ramp up and develop their skills and provide them with the adequate level of challenge.

Keywords : Aggregation - Natural Language Processing - Data Extraction - Machine Learning


Freelance Software Engineer - January 2014 to Dec. 2014

As the lead software engineer, I was primarily working on the prototyping and development of a sampling profiler based around the perf subsystem of the linux kernel. This product is being used both internally by Numscale engineers, and deployed as a web service. This product provides both low-level insight on multiple key aspects of the an application behaviour, but also a synthetic, high level report on how to improve the application and what results can be achieved.

Given my expertise of Image Processing and Computer Vision, I was also involved, as a consultant, in projects requiring those skills.

Keywords : High Performance Computing - Parallel Computing - SIMD - Systems - Profiling

DxO Labs

Image Processing R&D Engineer - July 2011 to Sept. 2013

As an Image Processing Engineer, I took part in the definition and implementation of image processing features (tonemapping, denoising,.) of the raw converter software DxO Optics Pro. As a framework engineer, I took part in the design and the implementation of the API provided to GUI teams, as well as the overall performance optimization of the correction engine.

Most notably, I designed and implemented a new HDR rendering and tone-mapping method that was highlighted as one of the prominent features of DxO Optics Pro 8. This work involved an extensive R\&D phase, based on the image quality evaluation of the previous method and its limitations. I also worked with various other methods of the image processing pipeline, such as distortion correction, denoising or raw conversion.

I also improved the performance of various component, in order to provide Optics Pro users with the best possible results. This was done using SIMD instructions, or OpenCL, as well as reducing the memory footprint of specific algorithms.

Keywords : Image Processing - Image Quality Evaluation - Image Quality Enhancement - Raw Pipeline - Photography

Universite Paris-Sud XI

Image Processing PhD Student - Sept. 2008 to July 2011 (public defense 14 Oct 2011)

The main objective of this work was to design and implement new ways to establish a collaboration between stereo-vision and motion analysis, in order to improve road safety and intelligent vehicles autonomy. The primary axis of this work was to combine stereo-vision and points correspondences in order to measure the sensor’s ego-motion. This ego-motion was finally compensated in order to detect independent motion.

This was implemented as a complete, real-time system that was embedded in was of the available test vehicles. The project was based on both proven components (through the OpenCV library) and original algorithm and made extensive use of modern parallel computing techniques such as SIMD and GPGPU (Cuda).

Finally, a fine study of imprecision sources was also conducted, which led to a set of recommandations, useful for every designer of intelligent and mobile vision system.

Keywords : Visual Odometry - Computer Vision - Intelligent Vehicles - Stereovision - Motion Analysis - Real-time

Stemmer Imaging

Field Application Engineer - Oct. 2007 to Sept.2008

As a Field Application Engineer, my first responsibility was to accompany industrial prospects and clients and help them design and implement a vision system meeting their requirements. I was involved in a wide variety of projects, ranging from industrial vision (quality control) to academic research (stereo vision) and to entertainment (real time motion capture)

Keywords : Machine Vision - Image processing - Acquistion Systems


PhD - Computer Vision (2008 - 2011)

Université Paris Sud & IFSTTAR

Master’s Degree - Engineer Diploma (2004 - 2007)

Institut d’Optique Graduate School

Physics/Optics, specialization in Image Processing

Intensive preparation to nation-wide competitive exams (2002 - 2004)

High School Diploma (2002)



  • C++
  • C
  • Java
  • Python
  • Javascript
  • rust
  • Haskell
  • Perl

Other Languages

French - Native
English - Fluent
Japanese - Beginner