Image Classification on 3-class reference dataset (val)
67Validation Accuracy (3-class)On-device (Paper 1)
Evaluation results
| Method | Links | |
|---|---|---|
| On-device (Paper 1)Training time to equiv. loss=∼9 min, Full collect–train–deploy cycle=∼30 min, Inference rate (device)=6.3 FPS, Heatmap FPS cost (device)=∼0.2 FPS, Weight file (.bin)=∼83 KB, Firmware flash time=∼2 min (Arduino IDE)2026.04 | 67 | |
| webmcu-vision-webTraining time to equiv. loss=∼1 min, Full collect–train–deploy cycle=<3 min, Inference rate (device)=6.3 FPS, Heatmap FPS cost (device)=∼0.2 FPS, Weight file (.bin)=∼83 KB, Firmware flash time=∼30 s2026.04 | 67 |