YOLO26-Detection / README.md
qaihm-bot's picture
v0.58.0
be1d946 verified
|
Raw
History Blame Contribute Delete
9.8 kB
---
library_name: pytorch
license: other
tags:
- real_time
- android
pipeline_tag: object-detection
---
![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yolo26_det/web-assets/model_demo.png)
# YOLO26-Detection: Optimized for Qualcomm Devices
Ultralytics YOLO26 is a machine learning model that predicts bounding boxes and classes of objects in an image.
This is based on the implementation of YOLO26-Detection found [here](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/detect).
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/v0.58.0/src/qai_hub_models/models/yolo26_det) 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/v0.58.0/src/qai_hub_models/models/yolo26_det) 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-Detection on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/yolo26_det) for usage instructions.
## Model Details
**Model Type:** Model_use_case.object_detection
**Model Stats:**
- Model checkpoint: YOLO26-N
- Input resolution: 640x640
- Number of parameters: 2.4M
- Model size (float): 9.2 MB
- Model size (w8a16): 3.2 MB
## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| YOLO26-Detection | ONNX | float | Snapdragon® X2 Elite | 4.417 ms | 5 - 5 MB | NPU
| YOLO26-Detection | ONNX | float | Snapdragon® X Elite | 9.329 ms | 9 - 9 MB | NPU
| YOLO26-Detection | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 6.251 ms | 5 - 207 MB | NPU
| YOLO26-Detection | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 16.73 ms | 5 - 221 MB | NPU
| YOLO26-Detection | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.033 ms | 0 - 143 MB | NPU
| YOLO26-Detection | ONNX | float | Qualcomm® QCS8450 | 16.73 ms | 5 - 221 MB | NPU
| YOLO26-Detection | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 10.468 ms | 5 - 7 MB | NPU
| YOLO26-Detection | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 4.024 ms | 2 - 178 MB | NPU
| YOLO26-Detection | ONNX | float | Snapdragon® 8 Elite Mobile | 4.235 ms | 0 - 176 MB | NPU
| YOLO26-Detection | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 4.235 ms | 0 - 176 MB | NPU
| YOLO26-Detection | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 9.329 ms | 9 - 9 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® X2 Elite | 2.44 ms | 2 - 2 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® X Elite | 5.804 ms | 2 - 2 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.499 ms | 2 - 223 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 7.556 ms | 3 - 225 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 17.421 ms | 2 - 5 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.482 ms | 2 - 116 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® QCS8450 | 7.556 ms | 3 - 225 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 7.149 ms | 2 - 5 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 6.763 ms | 3 - 199 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.303 ms | 0 - 86 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 26.256 ms | 3 - 206 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 2.581 ms | 1 - 82 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 6.763 ms | 3 - 199 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.581 ms | 1 - 82 MB | NPU
| YOLO26-Detection | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 5.804 ms | 2 - 2 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® X2 Elite | 2.735 ms | 5 - 5 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® X Elite | 4.829 ms | 5 - 5 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 3.359 ms | 0 - 177 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 8.882 ms | 4 - 196 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® QCS8275 | 13.124 ms | 1 - 157 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.504 ms | 5 - 6 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® SA8775P | 5.856 ms | 0 - 158 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® SA8650P | 5.856 ms | 0 - 158 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® SA8255P | 5.856 ms | 0 - 158 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® QCS8450 | 8.882 ms | 4 - 196 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 5.637 ms | 5 - 11 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.247 ms | 5 - 166 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® SA7255P | 13.124 ms | 1 - 157 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 2.619 ms | 0 - 154 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® SA8295P | 9.362 ms | 1 - 169 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 2.619 ms | 0 - 154 MB | NPU
| YOLO26-Detection | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 4.829 ms | 5 - 5 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 2.313 ms | 2 - 2 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® X Elite | 4.858 ms | 2 - 2 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 2.991 ms | 2 - 202 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 8.084 ms | 0 - 177 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.381 ms | 0 - 10 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® SA8775P | 5.036 ms | 0 - 178 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® SA8650P | 5.036 ms | 0 - 178 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® SA8255P | 5.036 ms | 0 - 178 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 4.771 ms | 2 - 6 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 5.307 ms | 2 - 184 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 1.938 ms | 2 - 188 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 19.777 ms | 2 - 181 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® SA7255P | 8.084 ms | 0 - 177 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 2.248 ms | 2 - 185 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 5.307 ms | 2 - 184 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.248 ms | 2 - 185 MB | NPU
| YOLO26-Detection | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 4.858 ms | 2 - 2 MB | NPU
| YOLO26-Detection | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 4.219 ms | 0 - 191 MB | NPU
| YOLO26-Detection | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 10.915 ms | 0 - 206 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® QCS8275 | 14.315 ms | 0 - 165 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.815 ms | 0 - 17 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® SA8775P | 7.12 ms | 0 - 170 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® SA8650P | 7.12 ms | 0 - 170 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® SA8255P | 7.12 ms | 0 - 170 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® QCS8450 | 10.915 ms | 0 - 206 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 6.772 ms | 0 - 20 MB | NPU
| YOLO26-Detection | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.701 ms | 0 - 169 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® SA7255P | 14.315 ms | 0 - 165 MB | NPU
| YOLO26-Detection | TFLITE | float | Snapdragon® 8 Elite Mobile | 3.133 ms | 0 - 161 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® SA8295P | 10.972 ms | 0 - 177 MB | NPU
| YOLO26-Detection | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 3.133 ms | 0 - 161 MB | NPU
## License
* The license for the original implementation of YOLO26-Detection can be found
[here](https://github.com/ultralytics/ultralytics/blob/main/LICENSE).
## References
* [Ultralytics YOLO26: NMS-Free Real-Time Object Detection for Edge Devices](https://docs.ultralytics.com/models/yolo26/)
* [Source Model Implementation](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/detect)
## 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).