Testing vision-toolkit#
We test the library with:
- unit tests: verify the inner working of the library works as expected.
- integration tests: test cases with a given input and an expected output.
- performance/regression tests: based on datasets documented by papers, we compute a score for the results and how long it took to get them.
We prefer integration tests over unit tests as they are easier to understand and maintain.
Test Datasets#
| Dataset Name | Description | Paper/Report | Data/Repo | labeled | status |
|---|---|---|---|---|---|
Hollywood2 |
Multiple viewers watch different short scenes from movies. Coordinates are 2D cartesian. | paper | repo | yes | OK (Segmentation) |
Zemblys |
Viewers look at a target moving on different points on a screen chosen randomly.) | replication report | repo | yes | OK (Segmentation) |
ETRA2019 |
8 viewers look at scenes while performing a task. Both eyes are tracked. IMPORTANT: Data are not labeled. Only trajectories are available. | website | data | no | Maybe for performance regression |