Seeing at the limit: diffraction, sampling, and what a camera can never see
Physics sets a hard limit on what any camera can resolve. A tour of diffraction, sampling, and photon noise, and what no algorithm can honestly recover.
Signals
Machines that understand
what they see.
We build computer vision that works outside the lab, on the hardware you actually ship. Detection, tracking, depth, segmentation, and rendering, tuned until they hold up in front of real users.
01
What we build
01
Perception pipelines
The perception layer itself, sized to the device it runs on: a phone, a kiosk, an embedded board, or a GPU you already own.
02
Augmented Reality
Photorealistic, real-time virtual try-on for eyewear, apparel, beauty, and jewelry, with compositing that stays convincing on a moving face in a live camera.
03
Visual search & recognition
Embedding-based visual search, style and attribute recognition, and systems that connect what a camera sees to what a catalog contains.
04
Precision & safety-critical vision
Marker tracking, alignment, and measurement for medical and industrial domains where sub-millimeter accuracy is the requirement.
02
Deployments Ashwin has led
CHANEL
Global AR Eyewear
Eyewear try-on for the luxury counter: frames tracked to the face in three dimensions, rendered to a standard the brand would sign off on.
60+ countries
Microsoft
Virtual Dressing Room
Clothing rendered onto a moving shopper, with cloth simulation driven by depth tracking so the fit reads as fabric rather than a decal.
Real-time CV
Disney
Magic Mirror
An augmented-reality attraction running in two theme parks, built to hold up under park traffic and stay in character for every guest.
2 parks · millions of interactions
Systems designed and led by Ashwin Rajendraprasad in prior roles.
03
Field notes
Hands-on: real-time face tracking in the browser, no build tools
Build a working 478-landmark face tracker in a single HTML file: MediaPipe from a CDN, getUserMedia, and a canvas overlay. No build tools, no server.
Production computer vision: why the demo is the easy 20%
A computer-vision demo is about 20% of the work. The other 80% (devices, latency, fidelity, monitoring without labels, failure) is where projects die.
More on computer vision
Object detection in the browser with Transformers.js (no server) Real-time hand & pose tracking in the browser (a MediaPipe follow-up) Shipping vision to the edge: quantization, distillation, and the latency budget When the camera is the model: vision-language models on the edge Gaussian splatting grows up: radiance fields for real products Depth from anything: monocular 3D and the road to spatial AI Generative vision: the new image models and what they change for CV teams Segment anything, now just say it: SAM 3 and open-vocabulary vision Real-time detection in the attention era: YOLO grows up Vision without labels: what DINOv3 means for teams with small datasets
04
Interactive demos
Coming soon
In-browser computer vision demos
Live hand and face tracking, AR try-on, and visual search, all running entirely in your browser. Check back, or write to us for an early look.
Have a vision problem?
Tell us what your system needs to see — we'll tell you whether it's feasible, and what it takes.
Start a conversation or email hello@cloudsignal.ai