Hardware Efficiency on Human Activity Recognition using WISDM
0.803Inference Latency (ms)SPECTRA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SPECTRAPrecision=INT82026.03 | 0.803 | 1,245.6 | 2.79 | 3.29 | 110.576 | 0.118 | — | |
| SPECTRAPrecision=FP322026.03 | 1.071 | 933.5 | 7.98 | 3.29 | 184.293 | 0.197 | — | |
| SPECTRAPrecision=INT82026.03 | 1.2 | 840.3 | — | — | 2,683,400 | 2,900 | 21.6 | |
| SPECTRAPrecision=FP162026.03 | 1.4 | 692 | — | — | 3,258,500 | 3,500 | 24.3 | |
| SPECTRAPrecision=FP322026.03 | 1.7 | 588.2 | — | — | 3,833,500 | 4,100 | 27 | |
| TinierHAR2026.03 | 2.6 | 384.6 | — | — | 5,797,000 | 6,200 | 54 | |
| TinyHAR2026.03 | 4.463 | 224 | 115.112 | 80.32 | 105,433.614 | 112.8 | — | |
| TinyHAR2026.03 | 5.2 | 192.3 | — | — | 13,838,000 | 14,800 | 121.5 | |
| DeepConvLSTM2026.03 | 9.863 | 101.4 | 741.22 | 5.04 | 145,061.423 | 155.2 | — | |
| DConvLSTM2026.03 | 11.4 | 87.7 | — | — | 27,021,500 | 28,900 | 762 |