Computer Vision

Computer vision covers image representations, classical and learned visual features, recognition, detection, segmentation, retrieval, and visual foundation models. Medical imaging is treated here as a specialized computer-vision subdomain rather than a separate top-level area: it uses the same representation, detection, segmentation, benchmarking, and domain-shift tools, but with clinical validation constraints.

Knowledge map

Image representation and features come first, then the recognition, detection, and segmentation tasks, then the backbones that power them, and finally applied and medical systems.

flowchart TD
  Rep[Image Representation and Features] --> Recog[Classification and Detection]
  Rep --> Seg[Segmentation and Metrics]
  Recog --> Backbones[CNNs and Vision Transformers]
  Seg --> Backbones
  Backbones --> Applied[OCR, Benchmarking, Domain Shift]
  Applied --> Medical[Medical Imaging]

Reading path

Read image foundations, then recognition and segmentation, then backbones, applied systems, and the medical subdomain.

1. Image Representation: pixels, channels, and tensor layout.
2. Classical Image Processing: filters and gradients before learned features.
3. Feature Extraction: hand-designed descriptors and what they capture.
4. Data Augmentation: label-consistent transforms that expand training data.
5. Image Classification: whole-image label prediction.
6. Object Detection: localizing and classifying objects with boxes.
7. Rotated Object Detection: oriented boxes for rotated objects.
8. Pose Estimation: locating keypoints and body structure.
9. Semantic Segmentation: per-pixel class labels.
10. Instance Segmentation: per-object masks.
11. Detection and Segmentation Metrics: AP, AR, IoU, Dice, panoptic quality, and boundary metrics.
12. CNN Architectures: convolutional backbones and receptive fields.
13. Vision Transformers: patch-token attention models for images.
14. Self-Supervised Visual Learning: pretraining without labels.
15. Content-Based Image Retrieval: nearest-neighbor search over image embeddings.
16. OCR Pipelines: detecting and reading text in images.
17. Document Image Analysis and Field Extraction: structured extraction from document images.
18. Model Benchmarking: comparing vision models fairly.
19. Domain Shift: accuracy loss when deployment data differs from training.
20. Synthetic Data: rendered or generated training images and their transfer gap.
21. Medical Image Analysis: vision under clinical validation constraints.
22. MRI Segmentation: volumetric lesion and organ masks.
23. MRI Classification: patient-level prediction and honest splits.

Connections

• Deep Learning supplies the CNN and transformer backbones.
• Video Understanding extends these methods across time, and Domain Applications uses them end to end.

Nav

Learning path — Computer vision

Image Representation →