The Hailo-8™ AI accelerator brings industry-leading neural processing throughput and power efficiency to support a wide range of AI applications. The table below demonstrates some of Hailo-8’s best-in-class capabilities in the foundational neural tasks of object detection and classification. It achieves very high FPS and power efficiency (FPS/W) across a number of industry-standard neural network models, while also offloading post-processing from the host processor it is working with.
| NN Model | Input Resolution | FPS | Power [W] | FPS/ W |
|---|---|---|---|---|
| Classification | ||||
| ResNet-50 v1 | 224×224 | 1,332 | 3.45 | 386 |
| MobileNet_v2_1.0 | 224×224 | 2,444 | 2.152 | 1,135 |
| Object Detection | ||||
| SSD_MobileNet_v1 | 300×300 | 1,055 | 2.2 | 479 |
| YOLOv5m | 640×640 | 218 | 4.6 | 47.3 |
| Segmentation | ||||
| stdc1 | 1024×1920 | 54 | 2.9 | 18.6 |
| Multi stream object detection (8 streams) | ||||
| YOLOv3 | 608×608 | 69 | 4.9 | 14 |