Computer Vision with Neural Networks
Bring the missing detail back
BOTIE
Examples, experiments and workspace controls
Connect a tool's inputs to see what it produces. You can try a branch before connecting it to Result.
Try your own inputs
Change one fine-detail value while keeping the coarse map fixed. Find where that detail appears in the combined output. Then change the coarse value and count how many upsampled positions respond.
Context for this challenge
Expand each coarse entry by a factor of two along both image dimensions.
We are trying to mark cell boundaries in a microscope image. A coarse decoder map has lost detail that an earlier encoder map still contains. Let's bring those two sources of information together at the same resolution. These tiny invented maps illustrate one connection; learning to mark cells requires images paired with labelled pixel masks.
Inspect the component calculation
Repeat pixels( Small map: Coarse cell features from the decoder Scale factor: 2 )
Examples and results
- Coarse cell features from the decoder
2 - Required result
2 2 2 2
Try a fresh construction
This clears this draft, including its undo history. Earlier successful attempts stay recorded. Build again without opening hints to record an independent attempt.
Stuck? Inspect a worked construction
The tools below form one possible solution. You can assemble it, follow its signals, and undo that change to return to your own construction.
Assumptions and test precision
This builds one skip connection. A full U-Net concatenates channels at matching spatial positions, then learns further convolutions; flattening here only makes the two feature groups easy to inspect. Numerical tests allow a difference of 0.001. Lists and structures must also have the right shape.
Your last tested pieces can start the full construction. An existing draft there keeps its own pieces.
A wider cell image needs aligned coarse features
This check uses the pieces from “Bring encoder detail beside the enlarged decoder map”.
BOTIE
- Cell detail from the encoder
1 0 0 1 0 1 1 0 - Coarse cell features from the decoder
2 5
Use a number, text, or an array matching the requested result. For a matrix, put each row in brackets. Saving your prediction runs your current construction on these inputs and checks the earlier examples too.
Build a construction in the workspace above first. This case will test those pieces.
Why does concatenating feature groups preserve information differently from adding them?
A full U-Net combines channels at matching positions; flattening here only exposes the two groups for inspection.
This builds one skip connection. A full U-Net concatenates channels at matching spatial positions, then learns further convolutions; flattening here only makes the two feature groups easy to inspect.
Work through the Python challenges →