README.md

---
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/rtmdet/web-assets/model_demo.png)

# RTMDet: Optimized for Qualcomm Devices

RTMDet is a highly efficient model for real-time object detection,capable of predicting both the bounding boxes and classes of objects within an image.It is highly optimized for real-time applications, making it reliable for industrial and commercial use

This is based on the implementation of RTMDet found [here](https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet).
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/rtmdet) 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/rtmdet) 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 [RTMDet on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/rtmdet) for usage instructions.


## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Model checkpoint: RTMDet Medium
- Input resolution: 640x640
- Number of parameters: 27.5M
- Model size (float): 105 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| RTMDet | ONNX | float | Snapdragon® X2 Elite | 8.097 ms | 5 - 5 MB | NPU
| RTMDet | ONNX | float | Snapdragon® X Elite | 15.379 ms | 51 - 51 MB | NPU
| RTMDet | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 11.157 ms | 3 - 233 MB | NPU
| RTMDet | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 45.676 ms | 5 - 299 MB | NPU
| RTMDet | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 14.707 ms | 5 - 224 MB | NPU
| RTMDet | ONNX | float | Qualcomm® QCS8450 | 45.676 ms | 5 - 299 MB | NPU
| RTMDet | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 22.885 ms | 5 - 12 MB | NPU
| RTMDet | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 15.379 ms | 51 - 51 MB | NPU
| RTMDet | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 9.29 ms | 3 - 187 MB | NPU
| RTMDet | ONNX | float | Snapdragon® 8 Elite Mobile | 9.29 ms | 3 - 187 MB | NPU
| RTMDet | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 6.033 ms | 2 - 192 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® X2 Elite | 5.757 ms | 2 - 2 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® X Elite | 14.194 ms | 27 - 27 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 8.857 ms | 3 - 356 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 18.681 ms | 3 - 358 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 64.074 ms | 2 - 5 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 13.531 ms | 2 - 5 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® QCS8450 | 18.681 ms | 3 - 358 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 14.281 ms | 2 - 5 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 14.194 ms | 27 - 27 MB | NPU
| RTMDet | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 6.175 ms | 1 - 297 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 6.175 ms | 1 - 297 MB | NPU
| RTMDet | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 5.111 ms | 1 - 324 MB | NPU
| RTMDet | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 20.007 ms | 0 - 305 MB | NPU
| RTMDet | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 51.172 ms | 2 - 367 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 105.943 ms | 1 - 224 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 28.927 ms | 0 - 3 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® SA8775P | 36.501 ms | 0 - 225 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® SA8650P | 36.501 ms | 0 - 225 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® SA8255P | 36.501 ms | 0 - 225 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® QCS8450 | 51.172 ms | 2 - 367 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 36.373 ms | 0 - 62 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 14.951 ms | 0 - 228 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® SA7255P | 105.943 ms | 1 - 224 MB | NPU
| RTMDet | TFLITE | float | Qualcomm® SA8295P | 46.064 ms | 0 - 283 MB | NPU
| RTMDet | TFLITE | float | Snapdragon® 8 Elite Mobile | 14.951 ms | 0 - 228 MB | NPU
| RTMDet | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 9.889 ms | 0 - 229 MB | NPU

## License
* The license for the original implementation of RTMDet can be found
  [here](https://github.com/open-mmlab/mmdetection/blob/3.x/LICENSE).

## References
* [RTMDet: An Empirical Study of Designing Real-Time Object Detectors](https://github.com/open-mmlab/mmdetection/blob/3.x/README.md)
* [Source Model Implementation](https://github.com/open-mmlab/mmdetection/tree/3.x/configs/rtmdet)

## 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).