---
library_name: pytorch
license: gpl-3.0
tags:
- real_time
- android
pipeline_tag: object-detection

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/wedetect/web-assets/model_demo.png)

# WeDetect: Optimized for Qualcomm Devices

WeDetect is a text-conditioned object detection model that uses a prompt-then-detect strategy. Text class names are reparameterized into the model weights before export, producing a fixed-vocabulary detector that runs efficiently on-device. The model uses LTRB (Left-Top-Right-Bottom) regression with multi-scale feature maps at strides 8, 16, and 32.

This is based on the implementation of WeDetect found [here](https://github.com/WeChatCV/WeDetect).
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/v0.62.2/src/qai_hub_models/models/wedetect) 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/v0.62.2/src/qai_hub_models/models/wedetect) 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 [WeDetect on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.62.2/src/qai_hub_models/models/wedetect) for usage instructions.


## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Input resolution: 640x640
- Model checkpoint: wedetect_tiny
- Model size (detector) (float): 143 MB
- Model size (detector) (mixed_with_float): 35.9 MB
- Model size (text_encoder) (float): 592 B
- Model size (text_encoder) (mixed_with_float): 13.3 KB
- Number of parameters (detector): 37.3M
- Number of parameters (text_encoder): 3.00

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| detector | ONNX | float | Snapdragon® X2 Elite | 10.367 ms | 5 - 5 MB | NPU
| detector | ONNX | float | Snapdragon® X Elite | 21.464 ms | 71 - 71 MB | NPU
| detector | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 15.79 ms | 0 - 330 MB | NPU
| detector | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 82.002 ms | 2 - 328 MB | NPU
| detector | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 35.024 ms | 5 - 13 MB | NPU
| detector | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 20.741 ms | 0 - 77 MB | NPU
| detector | ONNX | float | Qualcomm® QCS8450 | 82.002 ms | 2 - 328 MB | NPU
| detector | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 31.729 ms | 5 - 12 MB | NPU
| detector | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 21.464 ms | 71 - 71 MB | NPU
| detector | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 11.277 ms | 1 - 257 MB | NPU
| detector | ONNX | float | Snapdragon® 8 Elite Mobile | 11.277 ms | 1 - 257 MB | NPU
| detector | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 8.205 ms | 3 - 262 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® X2 Elite | 10.675 ms | 9 - 9 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® X Elite | 16.604 ms | 35 - 35 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 11.325 ms | 2 - 362 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® 8 Gen 1 Mobile | 25.246 ms | 1 - 363 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 15.636 ms | 6 - 14 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 16.036 ms | 0 - 44 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® QCS8450 | 25.246 ms | 1 - 363 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 15.612 ms | 8 - 15 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 16.604 ms | 35 - 35 MB | NPU
| detector | ONNX | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 11.259 ms | 7 - 315 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® 8 Elite Mobile | 11.259 ms | 7 - 315 MB | NPU
| detector | ONNX | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 9.54 ms | 5 - 316 MB | NPU
| detector | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 15.312 ms | 0 - 360 MB | NPU
| detector | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 81.626 ms | 5 - 375 MB | NPU
| detector | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 34.712 ms | 5 - 12 MB | NPU
| detector | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 20.06 ms | 5 - 8 MB | NPU
| detector | QNN_DLC | float | Qualcomm® SA8775P | 30.339 ms | 3 - 297 MB | NPU
| detector | QNN_DLC | float | Qualcomm® SA8650P | 30.339 ms | 3 - 297 MB | NPU
| detector | QNN_DLC | float | Qualcomm® SA8255P | 30.339 ms | 3 - 297 MB | NPU
| detector | QNN_DLC | float | Qualcomm® QCS8450 | 81.626 ms | 5 - 375 MB | NPU
| detector | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 30.823 ms | 5 - 11 MB | NPU
| detector | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 10.974 ms | 5 - 275 MB | NPU
| detector | QNN_DLC | float | Qualcomm® SA7255P | 116.277 ms | 1 - 292 MB | NPU
| detector | QNN_DLC | float | Qualcomm® SA8295P | 86.746 ms | 0 - 284 MB | NPU
| detector | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 10.974 ms | 5 - 275 MB | NPU
| detector | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 8.0 ms | 5 - 286 MB | NPU
| detector | QNN_DLC | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 13.038 ms | 0 - 358 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 18.202 ms | 5 - 12 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 18.075 ms | 5 - 38 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® SA8775P | 18.211 ms | 1 - 307 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® SA8650P | 18.211 ms | 1 - 307 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® SA8255P | 18.211 ms | 1 - 307 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 18.709 ms | 7 - 13 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 10.951 ms | 0 - 304 MB | NPU
| detector | QNN_DLC | mixed_with_float | Qualcomm® SA7255P | 43.579 ms | 1 - 306 MB | NPU
| detector | QNN_DLC | mixed_with_float | Snapdragon® 8 Elite Mobile | 10.951 ms | 0 - 304 MB | NPU
| detector | QNN_DLC | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 8.061 ms | 5 - 325 MB | NPU
| detector | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 16.101 ms | 0 - 345 MB | NPU
| detector | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 95.008 ms | 0 - 335 MB | NPU
| detector | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 35.407 ms | 0 - 83 MB | NPU
| detector | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 21.346 ms | 0 - 7 MB | NPU
| detector | TFLITE | float | Qualcomm® SA8775P | 31.601 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Qualcomm® SA8650P | 31.601 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Qualcomm® SA8255P | 31.601 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Qualcomm® QCS8450 | 95.008 ms | 0 - 335 MB | NPU
| detector | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 31.973 ms | 0 - 83 MB | NPU
| detector | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 11.789 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Qualcomm® SA7255P | 117.88 ms | 1 - 266 MB | NPU
| detector | TFLITE | float | Qualcomm® SA8295P | 83.175 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Snapdragon® 8 Elite Mobile | 11.789 ms | 0 - 267 MB | NPU
| detector | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 9.865 ms | 0 - 274 MB | NPU
| text_encoder | ONNX | float | Snapdragon® X2 Elite | 42.429 ms | 1 - 1 MB | NPU
| text_encoder | ONNX | float | Snapdragon® X Elite | 97.256 ms | 550 - 550 MB | NPU
| text_encoder | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 65.073 ms | 1 - 2267 MB | NPU
| text_encoder | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 147.311 ms | 1 - 1860 MB | NPU
| text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 102.652 ms | 1 - 4 MB | NPU
| text_encoder | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 93.386 ms | 0 - 644 MB | NPU
| text_encoder | ONNX | float | Qualcomm® QCS8450 | 147.311 ms | 1 - 1860 MB | NPU
| text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 99.381 ms | 1 - 3 MB | NPU
| text_encoder | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 97.256 ms | 550 - 550 MB | NPU
| text_encoder | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 46.947 ms | 1 - 2019 MB | NPU
| text_encoder | ONNX | float | Snapdragon® 8 Elite Mobile | 46.947 ms | 1 - 2019 MB | NPU
| text_encoder | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 39.926 ms | 1 - 2263 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® X2 Elite | 42.58 ms | 1 - 1 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® X Elite | 97.108 ms | 551 - 551 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 65.184 ms | 1 - 2267 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® 8 Gen 1 Mobile | 146.093 ms | 2 - 1862 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 102.648 ms | 1 - 4 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 97.242 ms | 0 - 643 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® QCS8450 | 146.093 ms | 2 - 1862 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 99.413 ms | 1 - 3 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® Dragonwing™ IQ-X7181 | 97.108 ms | 551 - 551 MB | NPU
| text_encoder | ONNX | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 47.98 ms | 1 - 2019 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® 8 Elite Mobile | 47.98 ms | 1 - 2019 MB | NPU
| text_encoder | ONNX | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 40.545 ms | 1 - 2263 MB | NPU
| text_encoder | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 58.729 ms | 0 - 2446 MB | NPU
| text_encoder | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 139.283 ms | 0 - 3011 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 95.311 ms | 0 - 3 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 82.142 ms | 0 - 2 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® SA8775P | 95.659 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® SA8650P | 95.659 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® SA8255P | 95.659 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® QCS8450 | 139.283 ms | 0 - 3011 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 93.468 ms | 0 - 2 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 45.558 ms | 0 - 1973 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® SA7255P | 260.988 ms | 1 - 2035 MB | NPU
| text_encoder | QNN_DLC | float | Qualcomm® SA8295P | 107.374 ms | 1 - 1903 MB | NPU
| text_encoder | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 45.558 ms | 0 - 1973 MB | NPU
| text_encoder | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 34.533 ms | 0 - 1983 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Snapdragon® 8 Gen 3 Mobile | 62.166 ms | 0 - 2447 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ IQ-8275 | 95.04 ms | 0 - 3 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 85.167 ms | 0 - 520 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® SA8775P | 95.579 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® SA8650P | 95.579 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® SA8255P | 95.579 ms | 1 - 2019 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ IQ-9075 | 93.403 ms | 0 - 2 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® Dragonwing™ Q-8750 | 46.559 ms | 0 - 1974 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Qualcomm® SA7255P | 261.126 ms | 1 - 2036 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Snapdragon® 8 Elite Mobile | 46.559 ms | 0 - 1974 MB | NPU
| text_encoder | QNN_DLC | mixed_with_float | Snapdragon® 8 Elite Gen 5 Mobile | 34.501 ms | 0 - 1982 MB | NPU
| text_encoder | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 73.275 ms | 1 - 3679 MB | NPU
| text_encoder | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 164.282 ms | 1 - 3564 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 109.282 ms | 0 - 553 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 99.84 ms | 1 - 4 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® SA8775P | 111.23 ms | 1 - 2145 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® SA8650P | 111.23 ms | 1 - 2145 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® SA8255P | 111.23 ms | 1 - 2145 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® QCS8450 | 164.282 ms | 1 - 3564 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 109.035 ms | 0 - 552 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 55.152 ms | 0 - 2074 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® SA7255P | 289.63 ms | 1 - 2145 MB | NPU
| text_encoder | TFLITE | float | Qualcomm® SA8295P | 112.884 ms | 1 - 2099 MB | NPU
| text_encoder | TFLITE | float | Snapdragon® 8 Elite Mobile | 55.152 ms | 0 - 2074 MB | NPU
| text_encoder | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 41.404 ms | 0 - 2220 MB | NPU

## License
* The license for the original implementation of WeDetect can be found
  [here](https://github.com/WeChatCV/WeDetect/blob/main/README.md).

## References
* [WeDetect: Fast Open-Vocabulary Object Detection as Retrieval](https://arxiv.org/abs/2512.12309)
* [Source Model Implementation](https://github.com/WeChatCV/WeDetect)

## 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).