Computer Vision Lab
From pixels to visual intelligence: try computer vision algorithms and their real models inside your browser, and understand what happens at each step.
- 46lessons
- 38experiments
- 56algorithms
- 10models running in the browser
Labs
- Image processingImage Processing LabGreyscale, brightness and contrast, filters, Sobel and Canny edges, thresholds and Otsu, morphology, CLAHE and histogram.
- Convolution and neural networksConvolution & CNN LabWatch the kernel pass over the image step by step, with padding, stride, pooling, ReLU and feature maps.
- Visual explanation and attention mapsExplainable AI & Visualization LabGrad-CAM, Grad-CAM++, Occlusion, LIME and Attention Rollout on real models, with an explanation of how they differ.
- Image classificationImage Classification LabClassify your image with officially trained ImageNet models, and read the top 5 probabilities and each model's card.
- Object detectionObject Detection LabDetect objects with the SSDlite model trained on COCO, with a confidence threshold, NMS and class hiding.
- Image segmentationImage Segmentation LabSemantic segmentation with LR-ASPP and DeepLabV3, with class colours, opacity and pixel counts.
- Medical imagesMedical Imaging LabDICOM and NIfTI, display windows, the three planes, registration and segmentation, and Dice and IoU metrics against a ground truth.
- Model comparisonModel Comparison LabCompare models on the same image: time, agreement and probabilities, without unfair comparisons.
Start here: a suggested path
- 1Digital image basics
- 2Colour spacesColor spaces: RGB, grayscale, HSV
- 3Introduction to image processing
- 4Histogram and contrast enhancement
- 5Spatial filtering and smoothing
- 6Edge detection
- 7Thresholding and classical segmentation
- 8Morphological processing
Experiments for beginners (16)
- Converting the image to greyscale and reading the histogram
- Brightness and contrast
- Revealing shadows with gamma correction
- Gaussian versus median
- The bilateral filter preserves edges
Experiments for researchers (11)
- Grad-CAM for different classes
- Grad-CAM++ versus Grad-CAM
- The effect of the occlusion patch size
- The stability of the LIME explanation
- Attention Rollout on DeiT
Medical imaging experiments (11)
- Preset CT windows
- The three planes in the phantom
- A 3D threshold and measuring Dice
- 3D region growing
- The effect of noise on segmentation
