Hugging Face's logo

STMicroelectronics
/
st_yolo_x

Object Detection
Model card Files
xet
Community
st_yolo_x / README.md
FBAGSTM's picture
FBAGSTM
Release AI-ModelZoo-4.0.0
e54bc05 verified
preview code
|
raw
history blame contribute delete
33.3 kB
metadata 
license: other
license_name: sla0044
license_link: >-
  https://github.st.com/AIS/stm32ai-modelzoo/raw/master/object_detection/LICENSE.md
pipeline_tag: object-detection

ST Yolo X quantized

Use case : Object detection

Model description

ST Yolo X is a real-time object detection model targeted for real-time processing implemented in Tensorflow. This is an optimized ST version of the well known yolo x, quantized in int8 format using tensorflow lite converter.

Network information

Network information Value
Framework TensorFlow Lite
Quantization int8
Provenance
Paper

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

Recommended Platforms

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

Performances

Metrics

Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.

Reference NPU memory footprint based on COCO Person dataset (see Accuracy for details on dataset)

Model Hyperparameters (depth_width) Dataset Format Resolution Series Internal RAM (KiB) External RAM (KiB) Weights Flash (KiB) STEdgeAI Core version
st_yoloxn d033_w025 COCO-Person Int8 192x192x3 STM32N6 333 0 877.66 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 256x256x3 STM32N6 624 0 884.91 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 320x320x3 STM32N6 1125 0 895.03 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 416x416x3 STM32N6 2676.12 0 904.03 3.0.0
st_yoloxn d050_w040 COCO-Person Int8 256x256x3 STM32N6 833 0 2414.64 3.0.0
st_yoloxn d100_w025 COCO-Person Int8 480x480x3 STM32N6 2707.5 0 1173.25 3.0.0
st_yoloxn d033_w025 ST-Person Int8 416x416x3 STM32N6 2676.12 0 906.28 3.0.0
st_yoloxn d067_w025 ST-Person Int8 416x416x3 STM32N6 2681.41 0 1039.78 3.0.0
st_yoloxn d100_w025 ST-Person Int8 416x416x3 STM32N6 2676.12 0 1173.28 3.0.0
st_yoloxn d100_w025 ST-Person Int8 480x480x3 STM32N6 2707.5 0 1173.25 3.0.0
st_yoloxn d100_w025 ST-Person W4A8 480x480x3 STM32N6 2593 0 738.53 3.0.0
st_yoloxn d100_w025 COCO-80-classes W4A8 480x480x3 STM32N6 3491.48 0 1217.59 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 192x192x3 STM32N6 360 0 519.06 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 256x256x3 STM32N6 624 0 526.31 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 320x320x3 STM32N6 1125 0 638.67 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 416x416x3 STM32N6 2670.84 0 575.53 3.0.0
st_yoloxn d050_w040 COCO-Person W4A8 256x256x3 STM32N6 835 0 1671.08 3.0.0
st_yoloxn d100_w025 COCO-Person W4A8 480x480x3 STM32N6 2707.5 0 868.8 3.0.0

Reference NPU inference time based on COCO Person dataset (see Accuracy for details on dataset)

Model Hyperparameters (depth_width) Dataset Format Resolution Board Execution Engine Inference time (ms) Inf / sec STEdgeAI Core version
st_yoloxn d033_w025 COCO-Person Int8 192x192x3 STM32N6570-DK NPU/MCU 6.63 150.74 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 256x256x3 STM32N6570-DK NPU/MCU 9.42 106.17 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 320x320x3 STM32N6570-DK NPU/MCU 13.29 75.26 3.0.0
st_yoloxn d033_w025 COCO-Person Int8 416x416x3 STM32N6570-DK NPU/MCU 21.3 46.95 3.0.0
st_yoloxn d050_w040 COCO-Person Int8 256x256x3 STM32N6570-DK NPU/MCU 20.12 49.70 3.0.0
st_yoloxn d100_w025 COCO-Person Int8 480x480x3 STM32N6570-DK NPU/MCU 35.8 27.93 3.0.0
st_yoloxn d033_w025 ST-Person Int8 416x416x3 STM32N6570-DK NPU/MCU 21.66 46.16 3.0.0
st_yoloxn d067_w025 ST-Person Int8 416x416x3 STM32N6570-DK NPU/MCU 24.37 41.02 3.0.0
st_yoloxn d100_w025 ST-Person Int8 416x416x3 STM32N6570-DK NPU/MCU 27.02 36.99 3.0.0
st_yoloxn d100_w025 ST-Person Int8 480x480x3 STM32N6570-DK NPU/MCU 35.8 27.93 3.0.0
st_yoloxn d100_w025 ST-Person W4A8 480x480x3 STM32N6570-DK NPU/MCU 34.27 29.18 3.0.0
st_yoloxn d100_w025 COCO-80-classes W4A8 480x480x3 STM32N6570-DK NPU/MCU 49.34 20.27 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 192x192x3 STM32N6570-DK NPU/MCU 6.33 157.96 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 256x256x3 STM32N6570-DK NPU/MCU 8.20 121.95 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 320x320x3 STM32N6570-DK NPU/MCU 11.82 84.66 3.0.0
st_yoloxn d033_w025 COCO-Person W4A8 416x416x3 STM32N6570-DK NPU/MCU 19.87 50.32 3.0.0
st_yoloxn d050_w040 COCO-Person W4A8 256x256x3 STM32N6570-DK NPU/MCU 18.34 54.52 3.0.0
st_yoloxn d100_w025 COCO-Person W4A8 480x480x3 STM32N6570-DK NPU/MCU 34.71 28.80 3.0.0

Reference MCU memory footprint based on COCO Person dataset (see Accuracy for details on dataset)

Model Hyperparameters (depth_width) Format Resolution Series Activation RAM (KiB) Runtime RAM (KiB) Weights Flash (KiB) Code Flash (KiB) Total RAM Total Flash STEdgeAI Core version
st_yoloxn d033_w025 Int8 192x192x3 STM32H7 184.92 12.54 891.18 108.38 197.46 999.56 3.0.0
st_yoloxn d033_w025 Int8 256x256x3 STM32H7 324.92 12.54 891.18 108.38 337.46 999.84 3.0.0
st_yoloxn d050_w040 Int8 256x256x3 STM32H7 451.4 15.63 2435.76 151.42 467.03 2587.18 3.0.0
st_yoloxn d033_w025 Int8 320x320x3 STM32H7 504.92 12.54 891.18 108.38 517.46 1000.04 3.0.0
st_yoloxn d033_w025 Int8 416x416x3 STM32H7 849.92 12.54 891.18 108.38 862.46 999.99 3.0.0

Reference MCU inference time based on COCO Person dataset (see Accuracy for details on dataset)

Model Hyperparameters (depth_width) Format Resolution Board Execution Engine Frequency Inference time (ms) STEdgeAI Core version
st_yoloxn d033_w025 Int8 192x192x3 STM32H747I-DISCO 1 CPU 400 MHz 357.79 3.0.0
st_yoloxn d033_w025 Int8 256x256x3 STM32H747I-DISCO 1 CPU 400 MHz 641.66 3.0.0
st_yoloxn d050_w040 Int8 256x256x3 STM32H747I-DISCO 1 CPU 400 MHz 1698.36 3.0.0
st_yoloxn d033_w025 Int8 320x320x3 STM32H747I-DISCO 1 CPU 400 MHz 1026.35 3.0.0
st_yoloxn d033_w025 Int8 416x416x3 STM32H747I-DISCO 1 CPU 400 MHz 1797.73 3.0.0

AP on COCO Person dataset

Dataset details: link , License CC BY 4.0 , Quotation[1] , Number of classes: 80, Number of images: 118,287

Model Hyperparameters (depth_width) Format Resolution Depth Multiplier Width Multiplier Anchors AP
st_yoloxn d033_w025 Int8 192x192x3 0.33 0.25 1 38.16 %
st_yoloxn d033_w025 Float 192x192x3 0.33 0.25 1 38.82 %
st_yoloxn d033_w025 W4A8 192x192x3 0.33 0.25 1 37.34 %
st_yoloxn d033_w025 Int8 256x256x3 0.33 0.25 1 44.85 %
st_yoloxn d033_w025 Float 256x256x3 0.33 0.25 1 45.12 %
st_yoloxn d033_w025 W4A8 192x192x3 0.33 0.25 1 44.43 %
st_yoloxn d050_w040 Int8 256x256x3 0.5 0.4 1 50.05 %
st_yoloxn d050_w040 Float 256x256x3 0.5 0.4 1 51.07 %
st_yoloxn d050_w040 W4A8 256x256x3 0.33 0.25 1 49.2 %
st_yoloxn d033_w025 Int8 320x320x3 0.33 0.25 1 51.17 %
st_yoloxn d033_w025 Float 320x320x3 0.33 0.25 1 51.69 %
st_yoloxn d033_w025 W4A8 320x320x3 0.33 0.25 1 50.61 %
st_yoloxn d033_w025 Int8 416x416x3 0.33 0.25 1 54.8 %
st_yoloxn d033_w025 Float 416x416x3 0.33 0.25 1 54.82 %
st_yoloxn d033_w025 W4A8 416x416x3 0.33 0.25 1 53.97 %
st_yoloxn d100_w025 Int8 480x480x3 1.0 0.25 3 61.1 %
st_yoloxn d100_w025 Float 480x480x3 1.0 0.25 3 61.7 %
st_yoloxn d100_w025 W4A8 480x480x3 0.33 0.25 1 59.9 %
st_yoloxn d033_w025 Int8 416x416x3 0.33 0.25 1 42.58 %
st_yoloxn d033_w025 Float 416x416x3 0.33 0.25 1 44.28 %
st_yoloxn d067_w025 Int8 416x416x3 0.67 0.25 1 46.2 %
st_yoloxn d067_w025 Float 416x416x3 0.67 0.25 1 46.7 %
st_yoloxn d100_w025 Int8 416x416x3 1.0 0.25 1 47.27 %
st_yoloxn d100_w025 Float 416x416x3 1.0 0.25 1 48.14 %
st_yoloxn d100_w025 Int8 480x480x3 1.0 0.25 1 48.15 %
st_yoloxn d100_w025 Float 480x480x3 1.0 0.25 1 48.68 %
st_yoloxn d100_w025 W4A8 480x480x3 1.0 0.25 1 47.3 %
st_yoloxn d100_w025 Int8 480x480x3 1.0 0.25 1 34.3 %
st_yoloxn d100_w025 Float 480x480x3 1.0 0.25 1 35.7 %

* 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

References

[1] “Microsoft COCO: Common Objects in Context”. [Online]. Available: https://cocodataset.org/#download. @article{DBLP:journals/corr/LinMBHPRDZ14, author = {Tsung{-}Yi Lin and Michael Maire and Serge J. Belongie and Lubomir D. Bourdev and Ross B. Girshick and James Hays and Pietro Perona and Deva Ramanan and Piotr Doll{'{a} }r and C. Lawrence Zitnick}, title = {Microsoft {COCO:} Common Objects in Context}, journal = {CoRR}, volume = {abs/1405.0312}, year = {2014}, url = {http://arxiv.org/abs/1405.0312}, archivePrefix = {arXiv}, eprint = {1405.0312}, timestamp = {Mon, 13 Aug 2018 16:48:13 +0200}, biburl = {https://dblp.org/rec/bib/journals/corr/LinMBHPRDZ14}, bibsource = {dblp computer science bibliography, https://dblp.org} }