Tasks and the End-to-End Pipeline
The canonical task family and the stages every vision system moves through, with a first look at evaluation.
Key Terms
ClassificationDetectionSegmentationDepthTrackingEvaluation Metrics
Python Tools
OpenCVNumPyMatplotlib
Textbook Reference
Szeliski Ch.1
Topics Covered
- 1Task family: classification, detection, segmentation, depth, and tracking by input and output type
- 2Pipeline stages: acquisition, preprocessing, representation, inference, and decision
- 3Data, model, and evaluation as co-equal design concerns
- 4Evaluation preview: accuracy, precision and recall, IoU, and the role of held-out data
- 5Failure modes from illumination, occlusion, scale, and viewpoint
Learning Outcomes
- Classify vision tasks by their input and output signatures
- Trace information flow through an end-to-end vision pipeline
- Describe the evaluation measures attached to each core task
Lecture Infographics
1 image file
Lecture Infographics
Lecture PDFs
Tasks and the End-to-End Pipeline
The embedded PDF preview is disabled to reduce Vercel bandwidth usage. Open the lecture file directly in Google Drive for viewing, downloading, or presenting.