Human Activity Recognition on USC-HAD
72Macro F1STELLA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| STELLA2026.07 | 72 | 76.36 | |
| TCNet2026.04 | 70.21 | 76.21 | |
| rTsfNet2026.04 | 65.76 | 64.85 | |
| Triplet LSTMvariant=HTL-SB2025.12 | 62.8 | — | |
| Triplet LSTM (HTL-SB)2026.07 | 62.8 | — | |
| SensorLLM2026.04 | 61.2 | 62.6 | |
| SensorLLM2026.07 | 61.2 | 62.6 | |
| LightTS2026.04 | 59.01 | 67.78 | |
| RAG-HAR2025.12 | 58.63 | 57.2 | |
| RAG-HAR2026.07 | 58.63 | 57.2 | |
| RandomForest2026.04 | 57.61 | 67.39 | |
| Crossformer2026.04 | 57.23 | 67.21 | |
| FiLM2026.04 | 55.66 | 64.23 | |
| Transformer-liketype=architecture2025.12 | 55 | — | |
| Transformer-like architecture2026.07 | 55 | — | |
| mobileHART2026.04 | 54.54 | 59.43 | |
| TimesNet2026.04 | 54.19 | 60.79 | |
| Triplet LSTMmode=baseline2025.12 | 53.5 | — | |
| Triplet LSTM baseline2026.07 | 53.5 | — | |
| iTransformer2026.04 | 52.63 | 58.9 | |
| FEDformer2026.04 | 51.37 | 62.91 | |
| Transformer2026.04 | 49.21 | 54.42 | |
| Softmax LSTMmode=baseline2025.12 | 49 | — | |
| DeepConvLSTM2026.04 | 48.8 | 50.6 | |
| ICGNet2026.04 | 46.93 | 52 | |
| Informer2026.04 | 46.29 | 53.67 | |
| Pyraformer2026.04 | 46.26 | 51.2 | |
| DeepConvLSTM2025.12 | 46 | — | |
| DeepConvLSTM2026.07 | 46 | — | |
| MchCnnGRU2026.04 | 43.07 | 52.6 | |
| Reformer2026.04 | 40.33 | 45.69 | |
| LLM as VA2026.07 | 32.1 | 34.51 | |
| HARGPT2026.07 | 21 | 28.9 | |
| CNN-LSTM [11]reference=[11], variant=standard CNN-LSTM2026.04 | — | 90.91 | |
| CNN-LSTM [12]reference=[12], variant=standard CNN-LSTM2026.04 | — | 90.88 | |
| CNN+HCreference=[7], description=heterogeneous CNN with grouped kernels for recalibration2026.04 | — | 90.67 | |
| two-level network architecture with dual-stage feature fusionarchitecture=dual-stage 1D CNN, fusion=intermediate2026.04 | — | 94.4 |