Score detections at the edge of a camera frame
An out-of-frame detection can change IoU even when the visible pixels agree. Clip both boxes before evaluating their overlap. Clipping the predicted box gives four visible intersection pixels and a union of eight. IoU reaches the inclusive 0.5 threshold.
Run the framework with your own input
Install Blyx 0.2.1 and a Rust toolchain with a host linker, then run the versioned project. Each positional argument is a number or a comma-separated numerical vector. The CLI accepts up to 16 entries per vector. The imported module accepts longer inputs.
git clone --depth 1 --branch v0.1.0 https://github.com/Rahulchaube1/blyx-osprey.git
cd blyx-osprey
blyxc native-run project.blyx -- "-2,0,2,2" "0,0,4,2" 4 4 0.5API and return values
Returns [visible IoU, matched, visible intersection area]. Match requires positive overlap and a score at least equal to the threshold. Invalid input returns an empty vector.
| Argument | Accepted input |
|---|---|
| pred | Vector of: finite number |
| truth | Vector of: finite number |
| width | positive finite number |
| height | positive finite number |
| threshold | probability in [0,1] |
Use the libraries independently
The framework vendors the same library source published in these two repositories. Each module has its own package page, input contract and regression suite.
Try the program
Expected output
0.5 1 4
Practice
Change pred = [0, 0, 1, 1], truth = [1, 0, 2, 1], threshold = 0.0. Predict the result, then run the program.
Hint
Returns [visible IoU, matched, visible intersection area]. Match requires positive overlap and a score at least equal to the threshold.
Worked solution
The result vector is [0, 0, 0]. Compare its fields with the return layout in the API section.
Find
let pred = [-2, 0, 2, 2];Replace with
let pred = [0, 0, 1, 1];Find
let truth = [0, 0, 4, 2];Replace with
let truth = [1, 0, 2, 1];Find
let threshold = 0.5;Replace with
let threshold = 0.0;Output after the change
0 0 0
