Every year in the UK hundreds gather for the Festival of the Unexceptional. In usually damp surroundings the most unglamorous vehicles of a little-gilded age joust for the prize of most mediocre. Last year’s winner was a 1983 Datsun Sunny 1.5 GL, a car unloved by even its mother, and guaranteed to fail to excite even the excitable. Yet there is a charm to the event, and the nostalgia-seeped glasses through which it is viewed don’t see back as far as the weekends spent scraping grey filler onto the bodywork, or the hours spent by the side of the road wondering if the roadside assistance man would have any spare snacks on him.
It’s a world that seems far removed from the glittering here-and-now of deep learning and AI. So, the question is clearly - has Sighthound spent countless man-months of PhD time training deep learning models that can spot an Austin Allegro from a Morris Ital? The somewhat disappointing answer is no. But if you need to know whether you were just passed by a Dacia Duster or a SSanyong Rexton, we can help. It would be a brave soul who predicted that the Duster or Rexton, or indeed their stable mates the Sandero or Korando, will be headlining the Festival of the Unexceptional in a few decades’ time, but maybe not too brave. But if they do, you will be able to identify them by using Sighthound.
It isn’t that we set out to bring the common-or-garden to our vehicle recognition software. We are as guilty as any of using high-end cars to promote our software, as a glance at the three Jaguars and Range Rover Velar in our last blog post on the subject will attest. But we have branched out from where we started. Our original vehicle recognition computer vision models were trained on vehicles marketed for sale in North America since 1991. For most US customers in the fields of law enforcement, parking and highways, this covers the vast majority of what they need. As Sighthound has spread to Europe, we’ve been asked to fill in the gaps in vehicle make and model coverage. We already had the Audis, BMWs, Mercedes and Land Rovers that make their way to US and Canadian showrooms, but we have now begun the process of adding those vehicles sold in the EU but not in North America, from an Alfa Romeo Brera to a Vauxhall Zafira. We have started with 132 model year body styles currently on sale, and will work our way back through vehicles for sale over the last 20 years. Citroen, Fiat, Peugeot, Renault, Seat, Skoda, Smart and Suzuki are all now represented.
The new EU models are live now on the Sighthound Cloud API and in the Sentry offline SDK, and you are able to test sample images on the Cloud demo page: https://www.sighthound.com/products/cloud
If your use case involves building products that search for vehicles where the number plate is unknown, or comparing the plate to the make and model, get in touch. We’d love to hear how we can help. If you’re wondering if we have added this feature to our home surveillance software, Sighthound Video, so that you know when your husband’s Ssanyong Rodius is entering the driveway, I’m afraid the answer is no, not yet. In the meantime, you may want to reconsider your life choices.