File size: 6,765 Bytes
4a991ff
 
309c3d7
53a826b
309c3d7
4a991ff
 
 
1dba911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4a991ff
309c3d7
 
 
 
 
 
 
 
 
1dba911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
309c3d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
---
title: README
emoji: 
colorFrom: red
colorTo: gray
sdk: static
pinned: false
---
<style>
  .st-header-banner {
    width: 100%;
    height: 120px;
    margin: 0 0 12px 0;
    padding: 0 24px;
    box-sizing: border-box;
    background: #03234B;
    display: flex;
    align-items: center;
    justify-content: flex-start;
  }
  .st-logo-right {
    height: 72px;
    width: auto;
    object-fit: contain;
  }
  img.rotating-content {
    width: 300px;
    height: 150px;
    object-fit:cover;
    object-position:center;
  }
  .icon {
    width: 1.5em;
    height: 1.5em;
    vertical-align: -.7em;
    padding: .25em .5em .25em .25em;
  }
  ul.social {
    list-style: none;
    padding-left: 0;
  }
  ul.social li {
    padding-left: .5em;
    display: flex;
  }
  .architectureImage {
    border: 0.5px solid black;
  }
</style>
<div class="st-header-banner">
  <img src="assets/ST_logo_2024_white.png" alt="ST Logo" class="st-logo-right"/>
</div>



_**Innovating with edge AI on STM32 and Hugging Face.**_

STMicroelectronics is a global semiconductor leader pushing artificial intelligence down to the most resource-constrained microcontrollers. With the **STM32 AI ecosystem**, ST provides an end-to-end pipeline — from pre-trained models in the **Model Zoo** to bare-metal optimized deployment — enabling embedded developers to build intelligent applications without deep ML expertise.
Models are optimized, quantized and validated to run directly on ST Neural-ART but also Cortex-M4, M7, M85 and M33 cores.

---
## End-to-End AI Pipeline
```
+----------------------------+
|          EXPLORE           |
+----------------------------+
|     STM32 AI Model Zoo     |
+----------------------------+
              |
              v
+----------------------------+
|           TRAIN            |
+----------------------------+
|  STM32 AI Model Zoo        |
|  Services                  |
+----------------------------+
              |
              v
+----------------------------+
|     OPTIMIZE / QUANTIZE    |
+----------------------------+
|  STM32 AI Model Zoo        |
|  Services                  |
+----------------------------+
              |
              v
+----------------------------+
|     EVALUATE / PREDICT     |
+----------------------------+
|  STM32 AI Model Zoo        |
|  Services                  |
+----------------------------+
              |
              v
+----------------------------+
|         BENCHMARK          |
+----------------------------+
|  STM32Cube AI Studio       |
|  STM32 Developer Cloud     |
+----------------------------+
              |
              v
+----------------------------+
|          CONVERT           |
+----------------------------+
|  STM32Cube AI Studio       |
|  ST Edge AI Core           |
+----------------------------+
              |
              v
+----------------------------+
|           DEPLOY           |
+----------------------------+
|  STM32Cube ecosystem       |
|  (tools, middleware, BSP)  |
+----------------------------+
```

This diagram summarizes the typical STM32 edge AI workflow from model discovery to on-device deployment:

1. **Explore**: Start from the STM32 AI Model Zoo to browse available architectures, pretrained checkpoints, and application examples.
2. **Train**: Use Model Zoo Services to retrain an existing model or build a task-specific pipeline on your own dataset.
3. **Optimize / Quantize**: Reduce model size and compute cost so the network fits embedded constraints while preserving the best possible accuracy.
4. **Evaluate / Predict**: Validate accuracy, inspect predictions, and compare tradeoffs before moving to hardware execution.
5. **Benchmark**: Measure latency, memory footprint, and target compatibility with STM32Cube AI Studio and STM32 Developer Cloud.
6. **Convert**: Transform the trained model into STM32-ready artifacts using STM32Cube AI Studio and ST Edge AI Core.
7. **Deploy**: Integrate the generated code into the STM32Cube ecosystem, including firmware, middleware, and board support components.

In short, the flow shows how a model moves from selection and training to optimization, hardware validation, and final integration on STM32 devices.

## Build, Optimize and Deploy AI/ML on STM32

- **STM32 AI Model Zoo**: A GitHub collection of reference machine learning models optimized for STM32 microcontrollers.
- **Application-Oriented Model Library**: A large set of models ready for re-training across multiple use cases.
- **Pre-trained Models Across Frameworks**: Reference models variants available for PyTorch, TensorFlow, and ONNX workflows.
- **End-to-End Scripts & Services**: Tools to retrain, quantize, evaluate, and benchmark models on custom datasets, plus autogenerated application code examples via [stm32ai-modelzoo-services](https://github.com/STMicroelectronics/stm32ai-modelzoo-services/tree/main)
- **Fast Deployment + Full Customization**: Use pretrained categories for quick deployment, or apply transfer learning / full training from scratch on your own data.
- **Reference Performance Metrics**: Results provided on STM32 MCU, NPU, and MPU targets for both float and quantized models.
- **Expanded Framework Support**: Comprehensive PyTorch support complements TensorFlow and ONNX in unified end-to-end workflows (train, evaluate, quantize, benchmark, deploy).

---

##  Key Tools & Ecosystem

- **STEdgeAI Core**: Converts trained neural networks into optimized C code for STM32.
- **STM32 AI Model Zoo services**: This repository provide scripts and workflows to ease end-to-end AI model training and integration on ST devices. They offer a valuable foundation to add AI capabilities to STM32-based projects.
- **STM32 AI Model Zoo** The repository with a of reference pre-trained machine learning models optimized for STM32 microcontrollers generated thanks to the STM32 AI Model Zoo services.
- **Integration with Popular Frameworks**:
  - TensorFlow / Keras
  - PyTorch (via ONNX export)
  - ONNX Runtime pipelines

---

## Links
- **[STM32 AI Model Zoo services](https://github.com/STMicroelectronics/stm32ai-modelzoo-services/tree/main)**
- **[STEdgeAI Core](https://www.st.com/en/development-tools/stedgeai-core.html)**
- **[STM32 Developer Cloud](https://stm32ai-cs.st.com/home)**
- **[STM32AI Model Zoo](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main)**
- **[STM32AI Cube Studio](https://www.st.com/en/development-tools/stedgeai-cubeai.html)**
---

## 🤝 Contact & Contributions

- For technical questions: [ST EdgeAI Community](https://community.st.com/t5/edge-ai/bd-p/edge-ai)
- For issues or feature requests, use the **Issues** or **Discussions** tabs in the respective repos.
- Contributions and feedback on models, pipelines, and docs are welcome.