Inference Latency Benchmarking on ITS Runtime Benchmark Policy B
0.8Latency (ms)DeepSets
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeepSetsParams (M)=0.09, Batch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 0.8 | 5,719 | |
| TransformerParams (M)=4.08, Batch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 1.2 | 3,785 | |
| Picture CNN (U-Net)Params (M)=6.12, Batch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 1.95 | 2,342 | |
| Set TransformerParams (M)=0.54, Batch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 2.44 | 1,870 | |
| 1D CNN + TransformerParams (M)=8.32, Batch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 4.59 | 993 | |
| K-medoidsBatch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 18.18 | 251 | |
| GreedyBatch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 54.19 | 84 | |
| K-meansBatch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 105.88 | 43 | |
| NSGA-IIBatch size=1, Hardware=GPU, Inference type=end-to-end2026.02 | 4,556.1 | 1 |