“GoPro loves France: Episode 2”: after having told you in 2019 about the geniuses of the algos and the signal processing which give all their power to the actioncams of the Californian giant, let’s lift the veil today on another team of frenchies. Also located in Issy-les-Moulineaux – well, when there is no pandemic! – this team is at the heart of a major news item for the brand: the launch of Quik.
Announced a fortnight ago, Quik is both an online storage and sharing service for your media (photos, videos, edits), as well as a largely automated editing mobile application. True Grail for the brand’s consumer strategy, Quik allows GoPro to orient itself a little more towards the service, and to put in the hands of the general public a tool to revolutionize the editing, pet peeve of neophytes.
If the storage (which arrives in the coming months) is managed by an American division of GoPro, most of the assembly part is developed in France. It is indeed a team built on the foundations of the start-up Stupeflix, bought by GoPro in 2016, that the actioncam champion has relied on to develop his digital magic. A magic sauce capable of transforming all the videos on your smartphone, whether from a GoPro or not, into a really nice montage to watch. Small tour in the kitchen.
A digital kitchen
” We could actually compare our application to a form of cooking! “, Laughs Guillaume Oulès, former Orange Valley, who joined the GoPro adventure through Stupeflix. ” In this kitchen, the algorithms would be the cooks, the pieces of clips selected by our grinders would be the ingredients and the filters to render the seasoning! », He continues, man.
The goal of the Frenchies of GoPro. Serve a dish – a montage – of quality.
” But also to serve it very quickly », He specifies. ” People want their edit now, they hardly ever go back to their files later. Not everyone has professional fitters like at GoPro! “.
An express character which partly prevents analyzes of colorimetric continuity in the case of heterogeneous sources, which was anyway ” a real technical puzzle, especially with the variety of Android terminals “.
But this urgency still leaves the possibility for algorithms to perform small miracles.
” Our algos analyze the clips to determine the context, human emotions expressed, nature of voices, facial expressions and smiles, actions (stable or jump for example, NDR), etc. While taking advantage of any metadata embedded in the file (read further). », Explains Guillaume Oulès.
A feat made possible in particular by the ever-growing power of terminals.
” Advances in smartphone chips are huge, whether it’s raw power or computer vision. Many tasks can be completed in an instant », Enthuses our interlocutor.
Added to this is the very finesse of the code of engineers of the French division. It mixes cutting-edge techniques, proven algorithms and tools from fundamental research … made in France.
High-tech, low tech, French Tech
Asked about the use of neural processors of SoCs integrated into smartphones, these famous artificial intelligence chips that Apple, Qualcomm and other Samsung put forward with each new generation, Guillaume Oulès smiles.
” No need at the moment! We are obviously working on it, but the CPU / GPU performance levels of modern terminals are sufficient for our needs. On this side, we are satisfied with a machine learning poor “, He describes, adding that” our approach is one of simplicity. Our programs must be robust and work on all terminals. “
Forests of decision trees, heuristics, GoPro France does well with AI, but with proven tools that are based on cutting-edge research in audio … it too made in France !
” We have been working with IRCAM since 2013-2014 and are integrating software bricks from their Ircam Amplify program to perform audio analysis of video clips », Adds Guillaume Oulès.
Temporal analysis, voice detection, tempo of music files with qualification of the different parts, French algorithms are still working for GoPro.
And it is the French who have asked other French people to help them.
” We work hand in hand with Mikael Kraak’s team (in charge of the image processing part within the GP1 processor that we have already interviewed, NDR). This allowed us to keep some of the information that the processor retrieves from the image sensor and other sensors. They are used in particular for the interpretation of colors or stabilization, these data were “thrown” after the calculation. We specified with them what information we could use in the metadata of the video files », Remembers Guillaume Oulès.
Data that significantly accelerates Quik’s work in image content analysis. And, ultimately, editing.
Franco-algo & American-Art
” There are big names in the French GoPro team, in particular members of FFMPEG. If I had to summarize what the French team brings, it would be everything related to editing, media manipulation, AI and automation and sound analysis. », Describes Guillaume Oulès.
If the big algos and the ugly mathematics are cooked in France, it is much from the United States that the artistic tools such as music videos come from ” given to the community, we did not deposit any rights on it », Specifies Guillaume. Or Quik’s color filters – the ” moods “In their jargon” which were developed in cooperation with our Senior Creative Director at GoPro, Abe Kislevitz “.
A creative who had to ” learn to translate their creative know-how, particularly in terms of colors. What he does to the eye, he had to translate it into precise calibration profiles », Relates Guillaume Oulès.
In short: all the image renditions and the montage identities offered by Quik come from the know-how of the creative champion of GoPro.
So if you use Quik and your automatic editing smacks of “American class”, it’s partly thanks to French science. It deserves a cocorico.