On Target Validation
User-managed validation runs the model on a real edge device and publishes the per-image predictions plus a detailed performance trace back to EdgeFirst Studio for accuracy and timing analysis. It is the right choice when you need to know how the model actually behaves on the hardware it will deploy to — not on a cloud instance.
The on-target tool for this workflow is the EdgeFirst Profiler.
For an alternative cloud-only flow, see On Cloud Validation. The managed flow provisions an EC2 instance and runs the model there — fast to set up, but the latency numbers will not reflect the deployment target.
Choose your path
Both paths produce the same Studio validation session and the same set of accuracy charts. The difference is where the validation session is created:
| If you... | Read |
|---|---|
| Want to create the session in the Studio web UI, then run the profiler on the device | Validation from Studio |
| Want to browse training sessions inside the profiler TUI and create the session in place | Validation from the Profiler |
Installation and connection
| Topic | Read |
|---|---|
| Install the profiler on the target board | Installation guides (per target) |
| Sign in to EdgeFirst Studio | Connecting to EdgeFirst Studio |
| Five-minute first profiling run | Quick Start |
What the profiler produces
Every validation run publishes two artifacts to the Studio session:
predictions.parquet— per-image predictions in EdgeFirst Dataset Format. EdgeFirst Studio consumes them to compute mAP and mIoU.trace.pftrace— a detailed timing trace covering the entire pipeline (decode → preprocess → inference → postprocess → NMS), per-operator backend timing (ORT nodes, TFLite ops, Neutron ticks, TensorRT layers, Hailo contexts), and system-metric counters (CPU, memory, temperature).
The trace is rendered in Studio's trace viewer directly from the session card.
Comparing runs
Two completed sessions can be opened side-by-side from the Model Experiments dashboard. See Validation Sessions for the comparison view.
Next steps
Once you have validated your model, see model deployment for the deployment paths to EdgeFirst Studio, the PC, embedded targets, the Maivin, and the Raivin.