Lighting the Torchđ„: Quentin Mace Deep Dive into DeepLabCutâs đ Backend Evolution
Meet Quentin Mace, an emerging talent from CentraleSupélec at Paris-Saclay University. Quentin recently wrapped up a productive six-month internship at the Mathis Lab, where he dedicated his efforts to advancing DeepLabCut.
What sparked𧚠your passion for engineering?
My fascination with mathematics đą began early on, sparked by a remarkable teacher in elementary school. This affinity for numbers grew into a deeper appreciation for science, particularly the rigor of the scientific method. For me, math is more than just numbers; itâs the purest form of science where every theory stands on proven facts and builds upon the wisdom of those before. The beauty lies in its clarity and precision.
As I ventured into engineering school, I delved deeper into advanced math and computer science. With each challenge, my passion grew stronger. Tutoring young students, I saw the broader impact of my knowledge. I realized the power of simplifying the complex, breaking down barriers, and making science universally accessible.
What made you come to DeepLabCut đ?
As I approached the end of my undergrad in engineering, a six-month internship was the next step. My goal? To dive deep into a project at the crossroads of AI and computer science, particularly one with real-world impact. DeepLabCut stood out â a project championing user-friendliness, empowering even non-coders to unlock the potential of computer vision in scientific research. Plus, some rave reviews from classmates familiar with the lab gave me the extra push.
What are the problems youâre focused on solving here at DLC đ§?
Computer vision research is continually evolving, with numerous innovative papers and updated libraries emerging each year. Given this dynamic landscape, PyTorch has grown increasingly popular as a deep learning framework. Recognizing this shift, DLC made the decision to integrate PyTorch while still providing support for Tensorflow. And I was right in the centre of this transformative journey.
Diving deep into the transition, my first mission was grounding the PyTorch version. Then, the real adventure began. I delved into the latest research, integrating state-of-the-art models, some of which performed brilliantly and have secured their spot in the upcoming release. Despite a few hitches, moments like achieving impeccable video tracking were sheer triumphs.
What Iâve found most rewarding about DeepLabCut is its intuitive user interface. Beneath its accessible exterior lies a sophisticated system. As users make the move from Tensorflow to PyTorch, theyâll likely appreciate the subtle improvements without confronting major changes.
What lessons learned and advice do you take with youâš?
The heart and soul of DeepLabCut is undoubtedly its incredible team. Surrounded by a dedicated team, I found support and collaborative spirit. It was a learning curve for me, both in terms of technical skills and the nuances of research.
My Python skills got better, I delved into different AI and computer vision techniques, and Iâm taking away a lot that will help as I look towards a PhDđȘ.
Oh, and on a lighter note, letâs not forget the boost to my Rubikâs cube game! đ€
What comes next?
The next chapter awaits back at the university, where Iâll be wrapping up my studies. I want to graduate soon and continue with my PhDđ. I see myself working in research for the next years.
Where do you find inspiration?
Ever stumbled upon 3Blue1Brown on Youtube? Itâs a channel that brings mathematics to life with stunning visuals and artistry. If you havenât, itâs worth the click â a true testament to the beauty of math!
