Home▸AI & Machine Learning▸Computer Vision

👁️ Computer Vision

A 3D pipeline showing how a convolutional neural network scans an image, builds a feature map, pools it down, and flags a detection region.

AI & Machine Learning3DAdvanced60 FPS

Watch a convolutional filter scan across a synthetic image in 3D, building a feature map behind it, pooling that map down, and drawing a detection box around the strongest activation.

🔬 What It Demonstrates

Each feature-map cell is the weighted sum of pixels under a sliding kernel window, passed through a ReLU. Max pooling then keeps only the strongest response in every 2×2 block, shrinking the map while preserving the signal.

🎮 How to Use

Change the kernel size, stride, filter type and image noise and watch the feature map and pooled map recompute live. Toggle the detection box to see the strongest region projected back onto the input image.

💡 Did You Know?

Yann LeCun's LeNet-5 (1998) used this exact convolution → pooling recipe to read handwritten digits — the same core building block still underlies modern object detectors like YOLO and Faster R-CNN.

⚙ Under the hood

A 3D pipeline showing how a convolutional neural network scans an image, builds a feature map, pools it down, and flags a detection region.

deep learningcomputer visionneural networksimage processingconvolutional neural networksfeature extractionThree.js

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install