Hi @amadahmads

Thanks for the question - and you’re right to call out that “a few seconds” can mean very different things.

For this challenge, please interpret the requirement as: results should be returned ideally in under 1 minute per assessment on a typical mid-range Android device (offline).

In practice, most usable solutions may likely fall in the 15–60 second range, but we recognise that:

• Higher-capacity models may reasonably take longer
• Field assessments are not always ultra time-critical
• There can be a trade-off between inference speed and robustness

What matters most is that:

• The app is practically usable in field conditions
• The workflow remains smooth and reliable
• The latency is justified by the quality and stability of results

We are not benchmarking pure model latency. Instead, we are assessing whether the solution is viable as a real-world offline assessment tool.