Testing deep learning neural networks on public datasets is fun, but it's usually on unseen data that you can really see how the published techniques really perform. Recently, I was trying to detect human faces on Game of Thrones footage. I was surprised to see that the most widely used techniques didn't fare really well. … Continue reading State of the art (2019) face detection with RetinaFace and MXNet
I've been lately regularly working with strategy consultancies, on data and AI matters. While corporate ambitions everywhere around data/AI are rising, when it comes to day-to-day operations, most of the data is still fragmented among many databases, and people use Excel sheets as exchange format. This in itself is interesting, because it means a corporate … Continue reading Excel to Pandas to PostgreSQL: Data Science for Strategy Consulting
AWS recently unveiled Outposts, a way to deploy on-premises servers that blend seamlessly in AWS cloud, and Snowball, a rugged server with a local EC2 and S3, that can be deployed anywhere on the planet even without connectivity. Apart from the engineering achievement, what’s interesting with those 2 announcements is that AWS seems to be … Continue reading AWS Outposts, Snowball: Precursors of AWS Region-as-a-service?
Every once in a while, you see an impressive tech arrive on the market. Usually the engineer in me is blown away by its elegance and/or the freshness of the technical approach to a problem. But my inner businessman wonders whether the company founders will manage to propose a compelling business case to their customer … Continue reading What Happened to Cloud Paging Technology?
A few years ago, I ran a cloud rendering startup that worked with VFX studios. We worked with a few mid-size studios, that used Windows on all their machines. We were faced with the challenge of deploying Windows server farms, and find practical ways to install programmatically heavy software (3dsMax, Maya, V-ray...) on all the … Continue reading Spoon: Containerization Before it was Cool, on Windows
After years of working with industry on large-scale numerical simulations, VFX studios on massive rendering, and other compute-intensive stuff such as AI-powered satellite image analysis, I have now 1 habit: I'm always looking for the biggest bang for the buck, preferably cloud-based since I hate operating hardware. When it comes to HPC (high-performance computing), Xeon … Continue reading Hijacking the Blade Shadow PC for Deep Learning
Once you're past the R&D phase, you have working deep neural nets, and you need to run them somewhere. If you want to avoid headaches with dependencies, you probably chose to make docker images out of your neural nets, with Tensorflow or equivalent + your code + your neural net weights. Problem is, if you … Continue reading Docker Registry Distribution in a Deep Learning Pipeline
Let's face it: nowadays big cloud providers have freed many companies from hardware. With cloud instances and object storage, CTOs don't have to worry anymore about hardware failures and redundancies. But the likes of Amazon, Azure etc did not stop there. Most providers also provide plenty of extra services (container orchestration, hosted database, "serverless" features … Continue reading Tech Independence: the Case for Excellence
Given a set of specs (specifications), an engineer usually tries to come up with the optimal architecture for those specs. As we all know, specs tend to change. The true understanding of a problem usually comes after you've started working on it. Like Mike Tyson used to say, everyone has a plan, until they get … Continue reading Good vs Bad Legacy Code
Over the years, I've used Jenkins, Concourse and a few other CI software. Recently, when the multistage dockerfile feature was released, it dawned on me that I used CI software for mainly 3 things: watching github repos, having a web UI to monitor builds, and be able to define pipelines. The last one is the … Continue reading Multistage Dockerfiles: do we still need CI Software?