Image Processing Inference on Swin T
3.38Average Latency (ms)ET QNN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ET QNNHardware=Samsung Galaxy S25 Ultra (Snapdragon 8 Elite), Backend=NPU, dtype=int82026.05 | 3.38 | 3.35 | 3.4 | |
| ET CoreMLHardware=Apple iPhone 15 Pro (A17 Pro), Backend=Auto, dtype=fp162026.05 | 8.7 | 8.4 | 9.13 | |
| ET XNNPACKHardware=Samsung Galaxy S25 Ultra (Snapdragon 8 Elite), Backend=CPU, dtype=int82026.05 | 23.06 | 21.06 | 26.36 | |
| ET VulkanHardware=Samsung Galaxy S25 Ultra (Snapdragon 8 Elite), Backend=GPU, dtype=fp162026.05 | 36.5 | 33.44 | 40.29 | |
| ET XNNPACKHardware=Google Pixel 9 Pro XL (Tensor G4), Backend=CPU, dtype=int82026.05 | 37.91 | 37.01 | 38.87 | |
| ONNXHardware=Google Pixel 9 Pro XL (Tensor G4), Backend=CPU, dtype=int82026.05 | 41.48 | 41.25 | 41.71 | |
| ONNXHardware=Samsung Galaxy S25 Ultra (Snapdragon 8 Elite), Backend=CPU, dtype=int82026.05 | 44.59 | 44.13 | 45.4 | |
| ET VulkanHardware=Google Pixel 9 Pro XL (Tensor G4), Backend=GPU, dtype=fp162026.05 | 98.86 | 93.38 | 102.59 |