Skip to content
All apps

CV Foundational Skills

About this app

The six foundational skills of computer vision live in the browser: classification (MobileNet), object detection (COCO-SSD), person segmentation (BodyPix), pose estimation (MoveNet), monocular depth estimation (principle demo) and OCR (Tesseract) — either on your own webcam or on generated logistics scenes, including two challenges on skill matching.

Six core computer vision tasks

Almost every computer vision application can be built from a few basic tasks. Classification answers what is in the image with one label for the whole picture. Object detection says where things are, with bounding boxes and classes. Segmentation assigns every pixel to a class. Pose estimation finds a person’s joints. Monocular depth estimation infers the distance of each pixel from a single image, using cues such as object size, occlusion and perspective. OCR reads text in the image. In practice these building blocks are combined to suit the question.

What the app shows

The app loads pretrained models directly into the browser: MobileNet for classification (1000 classes), COCO-SSD for detection (80 classes), BodyPix for person segmentation, MoveNet for pose with 17 keypoints and Tesseract for OCR. For depth it shows a hand-built illustrative depth map of the sample images, since no reliably small browser model is available. The source is your own webcam or logistics scenes (parcels, shelf, ramp). Everything runs locally and nothing is uploaded.

Try it yourself

Test the models on the drawn scenes and then on your webcam: on stylised images MobileNet and COCO-SSD often guess wrong, on real photos they work. That is the domain gap between training and deployment data. The two challenges pose practical cases, FEFO picking with a robot gripper and a ceiling camera above a warehouse. You choose the skills you need and get an explained solution.

Frequently asked questions

What is the difference between classification, object detection and segmentation?
Classification gives one label for the whole image. Object detection finds several objects and marks each with a box and class. Segmentation goes further and assigns every single pixel to a class, such as person or background.
What is pose estimation?
Pose estimation recognises a person’s posture by locating joints such as shoulders, elbows and knees. Models like MoveNet output 17 keypoints in COCO format. Uses include ergonomics and motion analysis.
What is the domain gap in computer vision?
A model works well on images similar to its training data and worse on others. A network trained on photos often misreads drawn scenes. New application areas therefore need suitable training data or fine-tuning.
Are webcam images uploaded by the app?
No. All models run locally in the browser, and neither webcam frames nor sample images leave the device. Only the model itself is downloaded the first time a skill is used.

Subject: Computer Vision

More from Computer Vision

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps