Performance Benchmark
We optimized AI models with Optimium and compared it with alternatives which are most widely used & known to be best for given target hardware. Check out the results below!
| Model | Target Hardware | Reference Tool | vs Reference |
|---|---|---|---|
| MobileNetV3 | AWS Graviton 3 | TensorFlow | 21.09x |
| ShuffleNetV2 | AWS Graviton 3 | TF Lite/XNNPACK | 2.30x |
| MobileNetV3 | AMD Ryzen9 7950x | Modular MAX Engine | 2.26x |
| MobileNetV3 | AMD Ryzen9 7950x | PyTorch | 3.26x |
| MobileNetV3 | AMD Ryzen9 7950x | TensorFlow | 9.6x |
| NasNet Mobile | AMD Ryzen9 7950x | Modular MAX Engine | 4.28x |
| NasNet Mobile | AMD Ryzen9 7950x | PyTorch | 5.07x |
| NasNet Mobile | AMD Ryzen9 7950x | TensorFlow | 16.84x |
| MediaPipe Pose Landmark(Lite) | RaspberryPi 5(Cortex-A76) | TF Lite/XNNPACK | 1.71x |
| MediaPipe Face Landmark | AMD Ryzen9 7950x | OpenVINO | 1.61x |
| MediaPipe Palm Detection(Full) | Qcom Kryo 585 Gold(Cortex-A77) | TF Lite/XNNPACK | 1.57x |
| MediaPipe Pose Landmark(Lite) | RaspberryPi 4(Cortex-A72) | TF Lite/XNNPACK | 1.57x |
| MediaPipe Palm Detection(Full) | Qcom Kryo 585 Gold(Cortex-A77) | TF Lite/XNNPACK | 1.55x |
| MediaPipe Pose Landmark(Lite) | RaspberryPi 4(Cortex-A72) | TF Lite/XNNPACK | 1.50x |
Updated about 1 year ago