Lightweight human detection models generated on high-quality human data sets. It can detect objects with high accuracy and speed in a total of 15 classes: Body, Male, Female, BodyWithWheelchair, BodyWithCrutches, Head, Face, Eye, Nose, Mouth, Ear, Hand, Hand-Left, Hand-Right, Foot. Even the classification problem is being attempted to be solved by object detection. There is no need to perform any complex affine transformations or other processing for pre-processing and post-processing of input images. In addition, the resistance to Motion Blur, Gaussian noise, contrast noise, backlighting, and halation is quite strong because it was trained only on images with added photometric noise for all images in the MS-COCO subset of the image set. In addition, about half of the image set was annotated by me with the aspect ratio of the original image substantially destroyed. I manually annotated all images in the dataset by myself. The model is intended to use real-world video for inference and has enhanced resistance to all kinds of noise. Probably stronger than any known model. However, the quality of the known data set and my data set are so different that an accurate comparison of accuracy is not possible.
This model does not use facial features, but only whole-body features to estimate gender. In other words, gender can be estimated even when the body is turned backwards and the face cannot be seen at all. This model is transfer learning using YOLOv9-Wholebody13 weights.
Don't be ruled by the curse of mAP.
Sample Score threshold >= 0.35 |
Sample Score threshold >= 0.35 |
|---|---|
 |
 |
 |
 |
 |
 |
output_wholebody15_e.mp4
-
The
gkey on the keyboard can be used to enable or disable the gender recognition mode. -
The
hkey on the keyboard can be used to enable or disable the hand recognition mode.output_wholebody15_e_switch.mp4
The use of CD-COCO: Complex Distorted COCO database for Scene-Context-Aware computer vision has also greatly improved resistance to various types of noise.
- Global distortions
- Noise
- Contrast
- Compression
- Photorealistic Rain
- Photorealistic Haze
- Motion-Blur
- Defocus-Blur
- Backlight illumination
- Local distortions
- Motion-Blur
- Defocus-Blur
- Backlight illumination
- COCO-Hand http://vision.cs.stonybrook.edu/~supreeth/COCO-Hand.zip
- CD-COCO: Complex Distorted COCO database for Scene-Context-Aware computer vision
- I am adding my own enhancement data to COCO-Hand and re-annotating all images. In other words, only COCO images were cited and no annotation data were cited.
- I have no plans to publish my own dataset.
Halfway compromises are never acceptable.
| Class Name | Class ID | Remarks |
|---|---|---|
| Body | 0 | Detection accuracy is higher than Male and Female bounding boxes. It is the sum of Male, and Female. |
| Male | 1 | Bounding box coordinates are shared with Body. It is defined as a subclass of Body as a superclass. |
| Female | 2 | Bounding box coordinates are shared with Body. It is defined as a subclass of Body as a superclass. |
| Body_with_Wheelchair | 3 | |
| Body_with_Crutches | 4 | |
| Head | 5 | |
| Face | 6 | |
| Eye | 7 | |
| Nose | 8 | |
| Mouth | 9 | |
| Ear | 10 | |
| Hand | 11 | Detection accuracy is higher than Hand_Left and Hand_Right bounding boxes. It is the sum of Hand_Left, and Hand_Right. |
| Hand_Left | 12 | Bounding box coordinates are shared with Hand. It is defined as a subclass of Hand as a superclass. |
| Hand_Right | 13 | Bounding box coordinates are shared with Hand. It is defined as a subclass of Hand as a superclass. |
| Foot (Feet) | 14 |
-
Python 3.10
-
onnx 1.16.1+
-
onnxruntime-gpu v1.18.1 (TensorRT Execution Provider Enabled Binary. See: onnxruntime-gpu v1.18.1 + CUDA 12.5 + TensorRT 10.2.0 build (RTX3070)
-
opencv-contrib-python 4.10.0.84+
-
numpy 1.24.3
-
TensorRT 10.2.0.19-1+cuda12.5
# Common ############################################ pip install opencv-contrib-python numpy onnx # For ONNX ########################################## pip uninstall onnxruntime onnxruntime-gpu pip install onnxruntime or pip install onnxruntime-gpu
-
Demonstration of models with built-in post-processing (Float32/Float16)
usage: demo_yolov9_onnx_wholebody15.py \ [-h] \ [-m MODEL] \ (-v VIDEO | -i IMAGES_DIR) \ [-ep {cpu,cuda,tensorrt}] \ [-it] \ [-dvw] \ [-dwk] \ [-dlr] \ [-dgm] options: -h, --help show this help message and exit -m MODEL, --model MODEL ONNX/TFLite file path for YOLOv9. -v VIDEO, --video VIDEO Video file path or camera index. -i IMAGES_DIR, --images_dir IMAGES_DIR jpg, png images folder path. -ep {cpu,cuda,tensorrt}, \ --execution_provider {cpu,cuda,tensorrt} Execution provider for ONNXRuntime. -it {fp16,int8}, --inference_type {fp16,int8} Inference type. Default: fp16 -dvw, --disable_video_writer Disable video writer. Eliminates the file I/O load associated with automatic recording to MP4. Devices that use a MicroSD card or similar for main storage can speed up overall processing. -dwk, --disable_waitKey Disable cv2.waitKey(). When you want to process a batch of still images, disable key-input wait and process them continuously. -dlr, --disable_left_and_right_hand_identification_mode Disable left and right hand identification mode. -dgm, --disable_gender_identification_mode Disable gender identification mode. -
YOLOv9-Wholebody15 - N - Swish/SiLU (PINTO original implementation, 2.4 MB)
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.632 0.467 0.501 0.320 body 2384 12642 0.702 0.636 0.690 0.478 male 2384 7035 0.606 0.579 0.566 0.434 female 2384 2897 0.371 0.390 0.331 0.248 body_with_wheelchair 2384 213 0.611 0.729 0.722 0.570 body_with_crutches 2384 112 0.522 0.848 0.775 0.616 head 2384 10719 0.806 0.710 0.765 0.525 face 2384 6152 0.810 0.632 0.689 0.428 eye 2384 5402 0.593 0.205 0.232 0.0881 nose 2384 5229 0.625 0.304 0.341 0.160 mouth 2384 4150 0.566 0.273 0.291 0.115 ear 2384 5005 0.626 0.314 0.350 0.171 hand 2384 8079 0.781 0.413 0.534 0.291 hand_left 2384 4061 0.637 0.285 0.401 0.223 hand_right 2384 4018 0.627 0.275 0.381 0.217 foot 2384 6909 0.596 0.411 0.451 0.233 -
YOLOv9-Wholebody15 - T - Swish/SiLU
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.758 0.583 0.646 0.450 body 2384 12642 0.793 0.756 0.808 0.615 male 2384 7035 0.688 0.678 0.699 0.588 female 2384 2897 0.549 0.545 0.555 0.461 body_with_wheelchair 2384 213 0.878 0.859 0.922 0.798 body_with_crutches 2384 112 0.775 0.875 0.904 0.833 head 2384 10719 0.862 0.797 0.848 0.612 face 2384 6152 0.855 0.744 0.802 0.545 eye 2384 5402 0.709 0.286 0.342 0.135 nose 2384 5229 0.749 0.413 0.473 0.244 mouth 2384 4150 0.680 0.368 0.413 0.180 ear 2384 5005 0.731 0.407 0.464 0.241 hand 2384 8079 0.886 0.558 0.703 0.426 hand_left 2384 4061 0.750 0.450 0.577 0.360 hand_right 2384 4018 0.756 0.437 0.561 0.352 foot 2384 6909 0.704 0.571 0.620 0.352 -
YOLOv9-Wholebody15 - T - ReLU
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.742 0.561 0.624 0.427 body 2384 12642 0.783 0.734 0.788 0.592 male 2384 7035 0.678 0.662 0.680 0.564 female 2384 2897 0.527 0.512 0.517 0.423 body_with_wheelchair 2384 213 0.826 0.822 0.891 0.766 body_with_crutches 2384 112 0.723 0.875 0.883 0.793 head 2384 10719 0.859 0.782 0.836 0.598 face 2384 6152 0.846 0.736 0.794 0.530 eye 2384 5402 0.684 0.268 0.324 0.124 nose 2384 5229 0.736 0.397 0.456 0.232 mouth 2384 4150 0.681 0.361 0.400 0.168 ear 2384 5005 0.727 0.383 0.440 0.227 hand 2384 8079 0.887 0.523 0.679 0.401 hand_left 2384 4061 0.744 0.414 0.549 0.335 hand_right 2384 4018 0.734 0.403 0.527 0.323 foot 2384 6909 0.694 0.539 0.590 0.327 -
YOLOv9-Wholebody15 - S - Swish/SiLU
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.823 0.651 0.724 0.531 body 2384 12642 0.853 0.808 0.856 0.699 male 2384 7035 0.762 0.721 0.771 0.683 female 2384 2897 0.674 0.639 0.686 0.603 body_with_wheelchair 2384 213 0.916 0.906 0.957 0.856 body_with_crutches 2384 112 0.850 0.911 0.919 0.886 head 2384 10719 0.886 0.843 0.888 0.668 face 2384 6152 0.893 0.779 0.833 0.623 eye 2384 5402 0.754 0.326 0.420 0.175 nose 2384 5229 0.828 0.474 0.557 0.311 mouth 2384 4150 0.769 0.427 0.501 0.239 ear 2384 5005 0.780 0.473 0.545 0.299 hand 2384 8079 0.918 0.660 0.794 0.525 hand_left 2384 4061 0.855 0.569 0.708 0.479 hand_right 2384 4018 0.837 0.566 0.701 0.469 foot 2384 6909 0.770 0.667 0.723 0.443 -
YOLOv9-Wholebody15 - C - Swish/SiLU
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.867 0.677 0.762 0.578 body 2384 12642 0.888 0.831 0.881 0.746 male 2384 7035 0.835 0.736 0.806 0.733 female 2384 2897 0.766 0.677 0.745 0.672 body_with_wheelchair 2384 213 0.895 0.925 0.968 0.880 body_with_crutches 2384 112 0.922 0.902 0.928 0.908 head 2384 10719 0.903 0.865 0.910 0.710 face 2384 6152 0.919 0.825 0.875 0.680 eye 2384 5402 0.806 0.364 0.458 0.198 nose 2384 5229 0.877 0.533 0.601 0.346 mouth 2384 4150 0.807 0.485 0.559 0.278 ear 2384 5005 0.826 0.523 0.601 0.343 hand 2384 8079 0.929 0.631 0.821 0.585 hand_left 2384 4061 0.910 0.571 0.754 0.544 hand_right 2384 4018 0.889 0.564 0.744 0.538 foot 2384 6909 0.825 0.717 0.781 0.506 -
YOLOv9-Wholebody15 - E - Swish/SiLU
Class Images Instances P R mAP50 mAP50-95 all 2384 82623 0.891 0.725 0.813 0.638 body 2384 12642 0.909 0.862 0.908 0.799 male 2384 7035 0.871 0.757 0.841 0.783 female 2384 2897 0.827 0.702 0.788 0.733 body_with_wheelchair 2384 213 0.928 0.962 0.983 0.904 body_with_crutches 2384 112 0.964 0.968 0.986 0.971 head 2384 10719 0.913 0.884 0.925 0.752 face 2384 6152 0.916 0.861 0.907 0.733 eye 2384 5402 0.829 0.441 0.556 0.257 nose 2384 5229 0.897 0.586 0.672 0.422 mouth 2384 4150 0.841 0.556 0.646 0.353 ear 2384 5005 0.854 0.593 0.676 0.409 hand 2384 8079 0.935 0.671 0.860 0.650 hand_left 2384 4061 0.924 0.624 0.800 0.610 hand_right 2384 4018 0.899 0.618 0.797 0.605 foot 2384 6909 0.856 0.782 0.842 0.585 -
Pre-Process
To ensure fair benchmark comparisons with YOLOX,
BGR to RGB conversion processingandnormalization by division by 255.0are added to the model input section. In addition, aresizing processfor input images has been added to improve operational flexibility. Thus, in any model, inferences can be made at any image size. The string1x3x{H}x{W}at the end of the file name does not indicate the input size of the image, but the processing resolution inside the model. Therefore, the smaller the values of{H}and{W}, the lower the computational cost and the faster the inference speed. Models with larger values of{H}and{W}increase the computational cost and decrease the inference speed. Since the concept is different from the resolution of an image, any size image can be batch processed. e.g. 240x320, 480x640, 720x1280, ...
-
Post-Process
Because I add my own post-processing to the end of the model, which can be inferred by TensorRT, CUDA, and CPU, the benchmarked inference speed is the end-to-end processing speed including all pre-processing and post-processing. EfficientNMS in TensorRT is very slow and should be offloaded to the CPU.
-
NMS default parameter
param value note max_output_boxes_per_class 20 Maximum number of outputs per class of one type. 20indicates that the maximum number of people detected is20, the maximum number of heads detected is20, and the maximum number of hands detected is20. The larger the number, the more people can be detected, but the inference speed slows down slightly due to the larger overhead of NMS processing by the CPU. In addition, as the number of elements in the final output tensor increases, the amount of information transferred between hardware increases, resulting in higher transfer costs on the hardware circuit. Therefore, it would be desirable to set the numerical size to the minimum necessary.iou_threshold 0.40 A value indicating the percentage of occlusion allowed for multiple bounding boxes of the same class. 0.40is excluded from the detection results if, for example, two bounding boxes overlap in more than 41% of the area. The smaller the value, the more occlusion is tolerated, but over-detection may increase.score_threshold 0.25 Bounding box confidence threshold. Specify in the range of 0.00to1.00. The larger the value, the stricter the filtering and the lower the NMS processing load, but in exchange, all but bounding boxes with high confidence values are excluded from detection. This is a parameter that has a very large percentage impact on NMS overhead. -
Change NMS parameters
Use PINTO0309/sam4onnx to rewrite the
NonMaxSuppressionparameter in the ONNX file.For example,
pip install onnxsim==0.4.33 \ && pip install -U simple-onnx-processing-tools \ && pip install -U onnx \ && python -m pip install -U onnx_graphsurgeon \ --index-url https://pypi.ngc.nvidia.com ### max_output_boxes_per_class ### Example of changing the maximum number of detections per class to 100. sam4onnx \ --op_name main01_nonmaxsuppression13 \ --input_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --output_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --input_constants main01_max_output_boxes_per_class int64 [100] ### iou_threshold ### Example of changing the allowable area of occlusion to 20%. sam4onnx \ --op_name main01_nonmaxsuppression13 \ --input_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --output_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --input_constants main01_iou_threshold float32 [0.20] ### score_threshold ### Example of changing the bounding box score threshold to 15%. sam4onnx \ --op_name main01_nonmaxsuppression13 \ --input_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --output_onnx_file_path yolov9_e_wholebody15_post_0145_1x3x480x640.onnx \ --input_constants main01_score_threshold float32 [0.15]
-
Post-processing structure
PyTorch alone cannot generate this post-processing. For operational flexibility,
EfficientNMSis not used.
-
-
INT8 quantization (YOLOv9-QAT)
If this work has contributed in any way to your research or business, I would be happy to be cited in your literature.
@software{YOLOv9-Wholebody15,
author={Katsuya Hyodo},
title={Lightweight human detection models generated on high-quality human data sets. It can detect objects with high accuracy and speed in a total of 15 classes: Body, Male, Female, BodyWithWheelchair, BodyWithCrutches, Head, Face, Eye, Nose, Mouth, Ear, Hand, Hand-Left, Hand-Right, Foot.},
url={https://github.com/PINTO0309/PINTO_model_zoo/tree/main/456_YOLOv9-Wholebody15},
year={2024},
month={8},
doi={10.5281/zenodo.10229410}
}I am very grateful for their excellent work.
-
COCO-Hand
https://vision.cs.stonybrook.edu/~supreeth/
@article{Hand-CNN, title={Contextual Attention for Hand Detection in the Wild}, author={Supreeth Narasimhaswamy and Zhengwei Wei and Yang Wang and Justin Zhang and Minh Hoai}, booktitle={International Conference on Computer Vision (ICCV)}, year={2019}, url={https://arxiv.org/pdf/1904.04882.pdf} }
-
CD-COCO: Complex Distorted COCO database for Scene-Context-Aware computer vision
@INPROCEEDINGS{10323035, author={Beghdadi, Ayman and Beghdadi, Azeddine and Mallem, Malik and Beji, Lotfi and Cheikh, Faouzi Alaya}, booktitle={2023 11th European Workshop on Visual Information Processing (EUVIP)}, title={CD-COCO: A Versatile Complex Distorted COCO Database for Scene-Context-Aware Computer Vision}, year={2023}, volume={}, number={}, pages={1-6}, doi={10.1109/EUVIP58404.2023.10323035} }
-
YOLOv9
https://github.com/WongKinYiu/yolov9
@article{wang2024yolov9, title={{YOLOv9}: Learning What You Want to Learn Using Programmable Gradient Information}, author={Wang, Chien-Yao and Liao, Hong-Yuan Mark}, booktitle={arXiv preprint arXiv:2402.13616}, year={2024} }
-
YOLOv9-QAT