Advanced Computer Vision

detection · segmentation · 3D vision

1:00

How modern computer vision systems locate, segment, and reconstruct objects — from YOLO bounding boxes to NeRF scene representations and vision transformers.

By the end, you can

Explain the four rungs of the vision ladder and what output each produces.
Contrast two-stage and one-stage detectors, naming an example of each and stating the accuracy-vs-speed tradeoff.
Describe how YOLO assigns grid-cell responsibility and what each cell outputs.
Explain what anchors are and why predicting offsets from anchors is easier than predicting raw coordinates.
Compute IoU given two axis-aligned bounding boxes and apply the standard acceptance threshold.
Trace the steps of Non-Maximum Suppression and explain why it is needed.
Distinguish semantic, instance, and panoptic segmentation.
Explain the role of skip connections in U-Net and how RoI Align improves Mask R-CNN.
Apply the stereo depth formula Z = (f · B) / d and interpret the relationship between disparity and depth.
Outline the three stages of Structure from Motion.
Describe what a NeRF represents and how it renders novel views via volume ray-marching.
Explain how ViT tokenizes an image and why it needs more pretraining data than a CNN.
Explain how DETR's object queries and Hungarian matching eliminate anchors and NMS.

Up next in AI, Machine Learning & Course Review