FIG. 007→
IoU, Dice, and Hausdorff: what each metric misses.
Overlap, boundary error, empty masks, and misleading averages. Build the metrics, inspect six exact cases, and design an evaluation that answers your task.
SegmentationEvaluation
LEARNING PATHS / THEORY → WORKING CODE
One idea builds
on the next.
Structured courses for understanding the machinery, one small, testable step at a time.
Start with the foundations, build a Transformer, then explore efficient inference and the systems around a model.
Explore the notebook/ FIND YOUR NEXT IDEA
Articles
ONE QUESTION, EXPLOREDFIG. 002→
DETR: object detection as set prediction.
What happens when we stop proposing boxes and start predicting a set of objects?
TransformersDetection
FIG. 003→
D-FINE: a better way to draw the box.
From four coordinates to fine-grained distributions. A closer look at localization refinement.
DetectionPaper notes
FIG. 004→
Linear attention: change the order.
A different kernel, a useful matrix identity, and a path around the quadratic attention matrix.
TheoryNumPy
FIG. 005→
The KV cache, without the mystery.
What a transformer remembers between tokens, and why faster decoding comes with a memory bill.
InferencePyTorch
FIG. 006→
An image is a sequence of patches.
Turning pixels into tokens, and following the shapes into a vision transformer.
VisionNumPy
Courses
CONNECTED LESSONS, A CLEAR DESTINATIONBASICSAVAILABLE · FREE
The foundations
Read equations and follow a learning step
- START WITH
- Python variables, lists, and functions
- THE PATH
- 4 published lessons · In sequence
Explore the course →Progress saved on this device
BASICSIN PROGRESS · FREE
Pixels → masks → evidence
Build and evaluate a reproducible binary mask
- START WITH
- Basic Python; arrays introduced as we go
- THE PATH
- 2 published lessons · In sequence
Explore the course →Progress saved on this device
INTERMEDIATEAVAILABLE · FREE
Attention & Transformers
Build a sequence-to-sequence model
- START WITH
- Foundations or equivalent
- THE PATH
- 8 published lessons · In sequence
Explore the course →Progress saved on this device
Lessons within courses
OPEN THE COURSE TO FOLLOW THE SEQUENCEThe foundations · LESSON 01What is actually inside a tensor?The foundations · LESSON 02How numbers mix information.The foundations · LESSON 03From scores to a distribution.The foundations · LESSON 04What actually changes when a model learns?Pixels → masks → evidence · LESSON 01Images are arrays. The contract matters.Pixels → masks → evidence · LESSON 02From pixels to masks: build a baseline you can question.Attention & Transformers · LESSON 01Words become vectors.Attention & Transformers · LESSON 02Give the sequence a sense of order.Attention & Transformers · LESSON 03Attention, from the ground up.Attention & Transformers · LESSON 04Who is allowed to see what?Attention & Transformers · LESSON 05Several heads, one shared representation.Attention & Transformers · LESSON 06The block around attention.Attention & Transformers · LESSON 07Encoder, decoder, or both?Attention & Transformers · LESSON 08Train it. Test what it learned.
No matches here. Try another topic or level, or explore the other content type.
6 articles · 3 available coursesKEEP ASKING ✳
On the drawing board 2 upcoming courses
These are outlines. Their lessons are not published yet.
Advanced attentionAdvanced · View the outline →AI systemsApplied systems · View the outline →