I don't get why startups collecting data in factories for rOboTiCs don't do this?
Realtime SOP checking is an obvious value-add for the factories they're already operating in. Make that the product, sell the data as an add-on (most of it is useless anyway 🗑️)
@sridharfyi@16vchq 1. A factory of small and specialised VLMs
2. Production CV pipelines are slow to build and break when conditions change. VLMs solve this but are locked in frontier labs and not edge friendly.
3. Team with years of domain expertise + starting up scar tissue
@16vchq We’re a vision factory: send us your video + objectives, get back a small VLM matching frontier quality on that task and ~10x cheaper
Our ICPs are video heavy teams that process videos at high volumes (OTT, surveillance, media houses etc)
Building it from india for the world 🇮🇳🚀
open sourcing Marlin-2B 🐟
a tiny VLM to extract structured information from videos
Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when?
Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params 🧵
@boulatbek Yes we’re the folks behind marlin 2B. And yes it does caption reels pretty well (although we did not have reels/tiktok data in the training mix specifically)