Computer vision memberikan persepsi visual ke sistem AI. Digunakan di robotika, autonomous vehicle, dan security.
Core Tasks
| Task | Description | Key Models |
|---|---|---|
| Classification | What object is this? | ResNet, EfficientNet |
| Object Detection | Where is the object? | YOLO, Faster R-CNN, DETR |
| Segmentation | Pixel-level classification | U-Net, Mask R-CNN |
| Pose Estimation | Human/keypoint detection | OpenPose, MediaPipe |
| OCR | Text extraction | Tesseract, TrOCR |
Modern Architectures
- CNN: convolution, pooling, batch norm
- Vision Transformer (ViT): patch embedding, attention
- YOLO: one-stage detection, anchor boxes
- Depth estimation: MiDaS, stereo matching
- 3D vision: NeRF, point cloud (PointNet++)
References
- YOLOv8: ultralytics
- OpenCV documentation
- Deep Learning for Computer Vision