Skip to main content

Advanced computer vision

From an image to a pipeline that counts vehicles and pedestrians at a crossroads. Ten modules, one running example, and every metric read from the COCO report you would receive in a real evaluation.

Course duration: 8h​

What you will learn​

  • Tell the three families of image tasks apart and pick the right annotation format
  • Train and compare a two-stage detector (Faster R-CNN) and a one-stage one (YOLOv8)
  • Set anchors, IoU and confidence thresholds without collapsing precision or recall
  • Read a COCO report, from IoU by hand to mAP at 0.5 and 0.5–0.95
  • Segment lanes and pavements with U-Net and DeepLab, then instances with Mask R-CNN
  • Track vehicles across frames with SORT and ByteTrack, and count them by a virtual line
  • Annotate, augment with Albumentations and detect near-duplicate leaks between train and test
  • Fine-tune YOLOv8 on a custom crossroads dataset and analyse errors by class and object size

Prerequisites​

  • Convolutional networks (course 10)
  • Python, NumPy and basic PyTorch

Course modules​

  1. The three families of image tasks
  2. Two-stage detection: the R-CNN family
  3. One-stage detection: YOLO and SSD
  4. Anchors, non-maximum suppression and thresholds
  5. Detection metrics: IoU and mean average precision
  6. Semantic segmentation: U-Net and DeepLab
  7. Instance segmentation with Mask R-CNN
  8. Multi-object tracking in video
  9. Annotation, augmentation and dataset quality
  10. Project: detection on a custom image dataset

Assessment and certificate​

The course ends with a 40-question exam covering every module. On success, a certificate of completion is issued; its number can be verified by anyone on the platform.

Free courses, by contrast, end with a 5-question quiz and a certificate preview, without certification.