Hugging Face's logo

qualcomm
/
YOLO26-Pose

Keypoint Detection
PyTorch
real_time
android
Model card Files
xet
 Community
YOLO26-Pose / README.md
qaihm-bot's picture
qaihm-bot
v0.59.0
e7dc855 verified
preview code
|
Raw
History Blame Contribute Delete
9.13 kB
---
library_name: pytorch
license: other
tags:
- real_time
- android
pipeline_tag: keypoint-detection
---
![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yolo26_pose/web-assets/model_demo.png)
# YOLO26-Pose: Optimized for Qualcomm Devices
Ultralytics YOLO26-Pose is a machine learning model that predicts human body keypoints and bounding boxes in an image.
This is based on the implementation of YOLO26-Pose found [here](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/pose).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/yolo26_pose) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
## Getting Started
Due to licensing restrictions, we cannot distribute pre-exported model assets for this model.
Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/yolo26_pose) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations
See our repository for [YOLO26-Pose on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/yolo26_pose) for usage instructions.
## Model Details
**Model Type:** Model_use_case.pose_estimation
**Model Stats:**
- Model checkpoint: YOLO26-N-Pose
- Input resolution: 640x640
- Number of parameters: 2.97M
- Model size (float): 11.4 MB
## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| YOLO26-Pose | ONNX | float | Snapdragon® X2 Elite | 5.365 ms | 9 - 9 MB | NPU
| YOLO26-Pose | ONNX | float | Snapdragon® X Elite | 10.666 ms | 11 - 11 MB | NPU
| YOLO26-Pose | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 6.899 ms | 9 - 214 MB | NPU
| YOLO26-Pose | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 17.948 ms | 8 - 230 MB | NPU
| YOLO26-Pose | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.916 ms | 0 - 13 MB | NPU
| YOLO26-Pose | ONNX | float | Qualcomm® QCS8450 | 17.948 ms | 8 - 230 MB | NPU
| YOLO26-Pose | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 11.863 ms | 8 - 11 MB | NPU
| YOLO26-Pose | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 10.666 ms | 11 - 11 MB | NPU
| YOLO26-Pose | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 4.728 ms | 5 - 185 MB | NPU
| YOLO26-Pose | ONNX | float | Snapdragon® 8 Elite Mobile | 4.728 ms | 5 - 185 MB | NPU
| YOLO26-Pose | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 4.398 ms | 5 - 187 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® X2 Elite | 2.642 ms | 4 - 4 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® X Elite | 6.595 ms | 3 - 3 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.829 ms | 0 - 231 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 7.964 ms | 4 - 236 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 18.822 ms | 3 - 6 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 6.295 ms | 2 - 6 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® QCS8450 | 7.964 ms | 4 - 236 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 6.84 ms | 3 - 6 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 6.595 ms | 3 - 3 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.834 ms | 0 - 91 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 2.834 ms | 0 - 91 MB | NPU
| YOLO26-Pose | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.48 ms | 2 - 95 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® X2 Elite | 3.377 ms | 5 - 5 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® X Elite | 6.326 ms | 5 - 5 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 4.149 ms | 5 - 198 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 10.228 ms | 4 - 216 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 16.721 ms | 0 - 167 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.771 ms | 5 - 7 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® SA8775P | 7.21 ms | 0 - 168 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® SA8650P | 7.21 ms | 0 - 168 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® SA8255P | 7.21 ms | 0 - 168 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® QCS8450 | 10.228 ms | 4 - 216 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 6.98 ms | 5 - 13 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 6.326 ms | 5 - 5 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 3.075 ms | 0 - 159 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® SA7255P | 16.721 ms | 0 - 167 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Qualcomm® SA8295P | 10.55 ms | 0 - 186 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 3.075 ms | 0 - 159 MB | NPU
| YOLO26-Pose | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.635 ms | 4 - 194 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 2.802 ms | 2 - 2 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® X Elite | 5.45 ms | 2 - 2 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.236 ms | 2 - 216 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 9.132 ms | 0 - 186 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.901 ms | 2 - 4 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® SA8775P | 5.655 ms | 0 - 185 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® SA8650P | 5.655 ms | 0 - 185 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® SA8255P | 5.655 ms | 0 - 185 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 5.398 ms | 1 - 6 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 5.45 ms | 2 - 2 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 22.883 ms | 2 - 188 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 6.039 ms | 3 - 197 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.591 ms | 2 - 187 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Qualcomm® SA7255P | 9.132 ms | 0 - 186 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 2.591 ms | 2 - 187 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.334 ms | 2 - 202 MB | NPU
| YOLO26-Pose | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 6.039 ms | 3 - 197 MB | NPU
| YOLO26-Pose | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.355 ms | 1 - 226 MB | NPU
| YOLO26-Pose | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 14.517 ms | 2 - 231 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 18.547 ms | 2 - 190 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.687 ms | 2 - 5 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® SA8775P | 9.177 ms | 2 - 191 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® SA8650P | 9.177 ms | 2 - 191 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® SA8255P | 9.177 ms | 2 - 191 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® QCS8450 | 14.517 ms | 2 - 231 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 8.882 ms | 2 - 25 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 3.855 ms | 0 - 181 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® SA7255P | 18.547 ms | 2 - 190 MB | NPU
| YOLO26-Pose | TFLITE | float | Qualcomm® SA8295P | 12.597 ms | 2 - 206 MB | NPU
| YOLO26-Pose | TFLITE | float | Snapdragon® 8 Elite Mobile | 3.855 ms | 0 - 181 MB | NPU
| YOLO26-Pose | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.262 ms | 0 - 190 MB | NPU
## License
* The license for the original implementation of YOLO26-Pose can be found
 [here](https://github.com/ultralytics/ultralytics/blob/main/LICENSE).
## References
* [Ultralytics YOLO26 Docs: Pose Estimation](https://docs.ultralytics.com/tasks/pose/)
* [Source Model Implementation](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/pose)
## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).