We organize the 5 different benchmarks as listed below. However, additional ones can be created upon request.
You are free to use grayscale, RGB and/or depth images or any other additional training data in order to perform the below benchmarks. The results will be marked accordingly to which data was used during training.
New results can be submitted using this form.
Possible variations
SVIRO can be used in additional other ways. For example:
- Train on infant seats with the handle down and test on infant seats with the handle up,
- Do not use everyday objects during training and verify the robustness when they are included during evaluation,
- Consider everyday objects and empty child/infant seats as an empty seat,
- Use a single or multiple cars to evaluate the generalization to new ones,
- Test segmentation if never a two (or three) seated car was seen,
- Train a single model for all seats, or one model per seat.
Let us know if you have further ideas on how to use our dataset!