| Applicable for | STM32MP13x lines, STM32MP15x lines, STM32MP21x lines, STM32MP23x lines, STM32MP25x lines |
1. Application samples[edit | edit source]↑
 |
 |
 | |
|---|---|---|---|
| NPU/GPU | CPU | CPU | |
 Image classification |
|
|
|
 Object detection |
|
|
|
 Pose estimation |
|
|
|
 Semantic segmentation |
|
|
|
 Face recognition |
|
|
|
 People tracking/heatmap |
|
|
|
 Important
|
| 1 NPU/GPU acceleration based on OpenVX backend using network binary graph (NBG) is available only for STM32MP257, STM32MP255 and STM32MP235 boards |
2. Performance Comparison[edit | edit source]↑
| Model | STM32MP25 NPU |
STM32MP15x CPU | |
|---|---|---|---|
 Image classification |
MobilenetV1_0.5 128x128 |
515 FPS |
22 FPS |
| MobilenetV2_1.0 224x224 |
72 FPS |
2 FPS | |
 Object detection |
SSD_MobilenetV1_1.0 300x300 |
40 FPS |
2 FPS |
| SSD_MobilenetV2_FPNLite 224x224 | 36 FPS | 0.8 FPS | |
 Face recognition |
Blazeface + Facenet |
33 FPS |
1.5 FPS |
 Pose estimation |
YoloV8n 256x256 |
59 FPS |
1.3 FPS |
 Semantic segmentation |
DeepLabV3 257x257 |
17 FPS |
1 FPS |
 People tracking/heatmap |
YoloV8n 320x320 |
51.8 FPS |
1.3 FPS |
Neural processing unit - NEW
Graphics processing units
Central processing unit
Frames per second
Neural processing unit - NEW
Graphics processing units
Central processing unit
Frames per second