Computer vision memberikan persepsi visual ke sistem AI. Digunakan di robotika, autonomous vehicle, dan security.

Core Tasks

TaskDescriptionKey Models
ClassificationWhat object is this?ResNet, EfficientNet
Object DetectionWhere is the object?YOLO, Faster R-CNN, DETR
SegmentationPixel-level classificationU-Net, Mask R-CNN
Pose EstimationHuman/keypoint detectionOpenPose, MediaPipe
OCRText extractionTesseract, 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

  1. YOLOv8: ultralytics
  2. OpenCV documentation
  3. Deep Learning for Computer Vision