Compare vision models for your task. Consider accuracy, response time, and hardware requirements in one view before choosing a backbone.
The largest model is not always the right model. Start with the task, the hardware available to your team, and the response time your application needs. A useful comparison makes those tradeoffs visible.
| Your priority | What to compare |
|---|---|
| Useful predictions | Errors on your real images |
| A responsive app | Latency on your hardware |
| A manageable footprint | Memory and compute needs |
Use the model catalog to explore possible starting points. Compare configurations side by side, then validate promising candidates against your own data. Catalog descriptions guide the choice; your evaluation should decide it.
Inspect candidate model configurations.
Make your selection understandable to the rest of the team. Record the assumptions behind it and the questions a training run should answer. This gives product and engineering teams a shared reason for the next experiment.
Choose a starting configuration.
Model catalog is part of the connected workflow: prepare, annotate, select, train, and deploy. Your team can revisit an earlier step when evaluation reveals something new. The product illustrations explain the workflow without uploading a dataset.