Hugging Face's logo

STMicroelectronics
/
st_yolodv2milli_pt

Object Detection
Model card Files
xet
Community
st_yolodv2milli_pt
File size: 13,289 Bytes 
cde7157
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
---
license: other
license_name: sla0044
license_link: >-
  https://github.com/STMicroelectronics/stm32ai-modelzoo/blob/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/LICENSE.md
pipeline_tag: object-detection
---
# **STYOLOMilli**

## **Use case** : `Object detection`

## **Model description**

STYOLOMilli is a **compact and efficient object detection model** designed for deployment on **resource‑constrained hardware** such as microcontroller units (MCUs) and neural processing units (NPUs). It is part of the STYOLO family introduced alongside STResNet in the paper [STResNet & STYOLO: A New Family of Compact Classification and Object Detection Models for MCUs](https://arxiv.org/abs/2601.05364).

The model builds on a highly compressed backbone derived from the STResNet design and integrates it within a YOLOX‑style architecture optimized for low memory usage, high efficiency, and competitive accuracy. STYOLOMilli delivers strong object detection performance (33.6 mAP on MS COCO) while maintaining a **very small model footprint**, outperforming or matching other micro sized detectors such as YOLOv5n and YOLOX‑Nano on the same tasks. 

The `st_yolo_milli_pt` variant is implemented in **PyTorch** and is tuned for **ultra‑efficient inference** on edge and MCU environments.

## **Network information**

| Network information | Value |
|--------------------|-------|
| Framework          | Torch |
| Quantization       | Int8 |
| Provenance         | [STMicroelectronics Model Zoo Services](https://github.com/STMicroelectronics/stm32ai-modelzoo-services/tree/main/object_detection/object_detection) |
| Paper              | [STResNet & STYOLO (arXiv:2601.05364)](https://arxiv.org/abs/2601.05364) |

The model is quantized to **int8** using **ONNX Runtime** and exported for efficient deployment.


## Network inputs / outputs

For an image resolution of NxM and NC classes

| Input Shape | Description |
| ----- | ----------- |
| (1, W, H, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |

| Output Shape | Description |
| ----- | ----------- |
| (1, (W/8xH/8 + W/16xH/16 + W/32xH/32), (NC+1+4))  | Model returns bounding boxes with 6 values for each box, four coordinates (x1,y1,x2,y2), class confidence and objectness confidence |


## Recommended Platforms

| Platform | Supported | Recommended |
|----------|-----------|-------------|
| STM32L0  | []        | []          |
| STM32L4  | []        | []          |
| STM32U5  | []        | []          |
| STM32H7  | []       | []          |
| STM32MP1 | []       | []          |
| STM32MP2 | []       | []         |
| STM32N6  | [x]       | [x]         |


# Performances

## Metrics

Measures are done with default STEdgeAI Core configuration with enabled input / output allocated option.

### Reference **NPU** memory footprint based on COCO dataset (see Accuracy for details on dataset)
| Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STEdgeAI Core version |
|-------|---------|--------|------------|--------|-------------------|-------------------|--------------------|-----------------------|
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_192/st_yolodv2milli_actrelu_pt_coco_192_qdq_int8.onnx) | COCO | Int8 | 192x192x3 | STM32N6 | 1008.00 | 0 | 3170.29 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_320/st_yolodv2milli_actrelu_pt_coco_320_qdq_int8.onnx) | COCO | Int8 | 320x320x3 | STM32N6 | 2744.48 | 800.00 | 3182.67 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_640/st_yolodv2milli_actrelu_pt_coco_640_qdq_int8.onnx) | COCO | Int8 | 640x640x3 | STM32N6 | 2768.00 | 9600.00 | 3189.38 | 3.0.0 |

* 640x640 coco checkpoints are provided primarily for finetuning purposes as these checkpoints are trained on large dataset at higher resolution. Models with 640 resolution is not suitable for deployment.

### Reference **NPU** inference time based on COCO dataset (see Accuracy for details on dataset)
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
|-------|---------|--------|------------|-------|------------------|--------------------|-----------|-----------------------|
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_192/st_yolodv2milli_actrelu_pt_coco_192_qdq_int8.onnx) | COCO | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 18.78 | 53.25 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_320/st_yolodv2milli_actrelu_pt_coco_320_qdq_int8.onnx) | COCO | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 70.57 | 14.17 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_640/st_yolodv2milli_actrelu_pt_coco_640_qdq_int8.onnx) | COCO | Int8 | 640x640x3 | STM32N6570-DK | NPU/MCU | 3417.54 | 0.29 | 3.0.0 |

### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
| Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STEdgeAI Core version |
|-------|---------|--------|------------|--------|-------------------|-------------------|--------------------|-----------------------|
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_192/st_yolodv2milli_actrelu_pt_coco_person_192_qdq_int8.onnx) | COCO-Person | Int8 | 192x192x3 | STM32N6 | 1008.00 | 0 | 3155.48 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_256/st_yolodv2milli_actrelu_pt_coco_person_256_qdq_int8.onnx) | COCO-Person | Int8 | 256x256x3 | STM32N6 | 2320.00 | 0 | 3164.48 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_320/st_yolodv2milli_actrelu_pt_coco_person_320_qdq_int8.onnx) | COCO-Person | Int8 | 320x320x3 | STM32N6 | 2743.88 | 800.00 | 3167.85 | 3.0.0 |

### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
|-------|---------|--------|------------|-------|------------------|--------------------|-----------|-----------------------|
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_192/st_yolodv2milli_actrelu_pt_coco_person_192_qdq_int8.onnx) | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 17.90 | 55.87 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_256/st_yolodv2milli_actrelu_pt_coco_person_256_qdq_int8.onnx) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 28.36 | 35.26 | 3.0.0 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_320/st_yolodv2milli_actrelu_pt_coco_person_320_qdq_int8.onnx) | COCO-Person | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 68.54 | 14.59 | 3.0.0 |



### AP on COCO dataset

Dataset details: [link](https://cocodataset.org/#download) , License [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode), Number of classes: 80

| Model | Format | Resolution | AP50 |
| --- | --- | --- | --- |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_192/st_yolodv2milli_actrelu_pt_coco_192.onnx) | Float | 3x192x192 | 35.42 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_192/st_yolodv2milli_actrelu_pt_coco_192_qdq_int8.onnx) | Int8 | 3x192x192 | 32.29 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_320/st_yolodv2milli_actrelu_pt_coco_320.onnx) | Float | 3x320x320 | 45.64 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_320/st_yolodv2milli_actrelu_pt_coco_320_qdq_int8.onnx) | Int8 | 3x320x320 | 43.79 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_640/st_yolodv2milli_actrelu_pt_coco_640.onnx) | Float | 3x640x640 | 52.64 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco/st_yolodv2milli_actrelu_pt_coco_640/st_yolodv2milli_actrelu_pt_coco_640_qdq_int8.onnx) | Int8 | 3x640x640 | 51.17 |

\* EVAL_IOU = 0.5, NMS_THRESH = 0.5, SCORE_THRESH = 0.001, MAX_DETECTIONS = 100

### AP on COCO-Person dataset

Dataset details: [link](https://cocodataset.org/#download) , License [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode) , Number of classes: 1


| Model | Format | Resolution | AP50 |
| --- | --- | --- | --- |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_192/st_yolodv2milli_actrelu_pt_coco_person_192.onnx) | Float | 3x192x192 | 61.71 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_192/st_yolodv2milli_actrelu_pt_coco_person_192_qdq_int8.onnx) | Int8 | 3x192x192 | 59.91 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_256/st_yolodv2milli_actrelu_pt_coco_person_256.onnx) | Float | 3x256x256 | 68.07 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_256/st_yolodv2milli_actrelu_pt_coco_person_256_qdq_int8.onnx) | Int8 | 3x256x256 | 66.16 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_320/st_yolodv2milli_actrelu_pt_coco_person_320.onnx) | Float | 3x320x320 | 72.12 |
| [st_yolodv2milli_pt](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yolodv2milli_pt/ST_pretrainedmodel_public_dataset/coco_person/st_yolodv2milli_actrelu_pt_coco_person_320/st_yolodv2milli_actrelu_pt_coco_person_320_qdq_int8.onnx) | Int8 | 3x320x320 | 70.91 |

\* EVAL_IOU = 0.5, NMS_THRESH = 0.5, SCORE_THRESH = 0.001, MAX_DETECTIONS = 100

## Retraining and Integration in a simple example:

Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)


## References


- **STYOLO / STResNet paper**  
  [S. Sah & R. Kumar, *STResNet & STYOLO: A New Family of Compact Classification and Object Detection Models for MCUs*](https://arxiv.org/abs/2601.05364)

- **YOLOX (inspires STYOLO architecture)**  
  [Ge et al., *YOLOX: Exceeding YOLO Series in 2021*](https://arxiv.org/abs/2107.08430)

- **MS COCO dataset**  
  [Lin et al., *Microsoft COCO: Common Objects in Context*](https://arxiv.org/abs/1405.0312)