Every lesson in the Computer Vision slide course, in full text: 2 decks, 168 slides.
Dataset Engineering for Instance Segmentation (PASTIS Parcels)A working-session deck for a hand-annotated PASTIS parcel instance-segmentation project, built around six findings measured from the student's own files: 83 instances per 128x128 tile, box sides of 4 to 38 px, and annotations byte-identical across all 61 observations of a tile. It covers round-trip QA of written artifacts, why per-observation duplication makes a random split leak an entire validation set, configuring Mask R-CNN anchors and min_size from measured object scale instead of COCO defaults, the dormant coordinate bug created by a duplicated 850 literal, polygons as source versus rasters as build output via COCO, fixed per-band normalisation, and a temporal-consistency filter that replaces a guessed pseudo-label confidence threshold with agreement across the time series.
Robust SIFT + RANSAC Landmark Matching (Before/After Registration)A PhD-level deck of 52 slides, about two hours long, on why SIFT and RANSAC landmark matching breaks down across before-and-after vitiligo skin photos, and how to fix it. It sets up a diagnosis-first framework - detect, describe, match, estimate - and then applies three principled upgrades to a real OpenCV pipeline: RootSIFT, built on the Hellinger kernel and the L1-normalize-then-square-root identity; a mutual nearest-neighbor cross-check layered on top of Lowe's ratio test, traded off as precision against recall for a robust estimator; and MAGSAC++ in place of vanilla RANSAC, for its sigma-consensus and threshold insensitivity. The deck includes five traps, five checks, a fully scaffolded your-turn implementation, and an end-to-end ablation. Every number and code path was executed against OpenCV 4.12 before the deck was written.
Want this taught 1-on-1? Alexander tutors Computer Vision — $55/session, free consultation.