develop/how_to/benchmark_a_model/

[giscus[bot]]

develop/how_to/benchmark_a_model/

A set of easy-to-use utilities that will come in handy in any computer vision project.

https://supervision.roboflow.com/develop/how_to/benchmark_a_model/

[David-rn]

Hi! Thanks for the great guide 🔥, it is quite easy to follow.

I haven't found anything on supervision, correct me if I'm wrong. Would it be possible to create something like a MetricSession?

For example, I'm testing three different models, but I want to combine all their benchmarks in a final plot, or in a table like the one in Roboflow's Model Leaderboard. The MetricsSession would be in charge of sorting the results from the different models that we input to ploting them afterwards.

[LinasKo]

Hi @David-rn,

Great to hear your thoughts!

No, indeed, there is no such concept at the moment, but you're right — aggregating and comparing metrics is the natural next step a user would take.

What does exist, is the to_pandas converters for each result. Stacking such dataframes should be fairly easy, but it'd be great if we had an automatic implementation.

I'll open an issue; we can discuss there.

[LinasKo]

Opened issue #1707.