M12 · Level II — Machine Intelligence
Computer Vision & Multimodal Perception
Derive visual perception from physical sensing and signal representation, then engineer deep vision and multimodal systems.
Topics
Technical concepts
- 01Image Formation: Scene -> Light -> Sensor
- 02Sampling, Quantisation and Aliasing
- 03Images as Arrays, Tensors and Coordinate Systems
- 04Filtering: Local Structure Before Learning
- 05Convolution and Cross-Correlation
- 06Receptive Fields and Hierarchical Features
- 07Convolutional Neural Networks
- 08Pooling, Striding and Downsampling
- 09Vision Data Normalisation and Augmentation
- 10Visual Classification
- 11Object Detection as Localisation + Recognition
- 12Matching, IoU and Non-Maximum Suppression
- 13Semantic Segmentation
- 14Instance and Panoptic Perception
- 15Multi-Scale Vision and Feature Pyramids
- 16Monocular Depth and Geometry
- 17Stereo Geometry
- 18Motion and Optical Flow
- 19Point Clouds and 3D Representation
- 20Vision Transformers
- 21CNN vs Vision Transformer: Inductive Bias
- 22Self-Supervised Visual Representation Learning
- 23Multimodal Perception: Shared State Across Modalities
- 24Image-Text Alignment
- 25Cross-Modal Attention and Grounding
- 26Document Vision and OCR as Perception
- 27Video Perception
- 28Uncertainty and Ambiguity in Visual Perception
- 29Vision Robustness and Distribution Shift
- 30Perturbation, Adversarial Sensitivity and Failure Surfaces
- 31Visual Attribution and Explanatory Probes
- 32Multimodal Failure: Alignment, Dominance and Shortcutting
- 33Quantum Technology Bridge: Vision, Kernels and Hybrid Perception
Practical implementation
- 01Laboratory I: Vision Operators from First Principles
- 02Laboratory II: CNN from Primitive Tensor Operations
- 03Laboratory III: Detection and Segmentation
- 04Laboratory IV: Stereo and Depth
- 05Laboratory V: Image-Text Retrieval and Grounding
- 06Laboratory VI: Vision Robustness Matrix
Tools and technologies
Project
End-to-end perception or multimodal system
Expected learning outcomes
- 01Can explain image formation, sampling and quantisation.
- 02Can implement basic vision operators without high-level black boxes.
- 03Can derive and audit CNN tensor transformations.
- 04Can evaluate detection and segmentation with task-appropriate metrics.
- 05Can reconstruct simple 3D geometry from stereo or depth.
- 06Can implement a small vision transformer and compare it with a CNN.
- 07Can build a measurable image-text alignment system.
- 08Can diagnose robustness, shift and multimodal shortcut failures.
- 09Can reproduce a vision/multimodal research claim with documented deviations.
Assessment
- 01Daily practice
- 02Laboratories
- 03Research reproduction
- 04Integrated project
- 05Technical viva
- 06Engineering record
Source document: TUTOR Frontier Module 12 Computer Vision Multimodal Perception