README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: object-detection

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/conditional_detr_resnet50/web-assets/model_demo.png)

# Conditional-DETR-ResNet50: Optimized for Qualcomm Devices

DETR is a machine learning model that can detect objects (trained on COCO dataset).

This is based on the implementation of Conditional-DETR-ResNet50 found [here](https://github.com/huggingface/transformers/tree/main/src/transformers/models/conditional_detr).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/conditional_detr_resnet50) 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
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/conditional_detr_resnet50/releases/v0.46.0/conditional_detr_resnet50-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/conditional_detr_resnet50/releases/v0.46.0/conditional_detr_resnet50-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/conditional_detr_resnet50/releases/v0.46.0/conditional_detr_resnet50-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[Conditional-DETR-ResNet50 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/conditional_detr_resnet50)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/conditional_detr_resnet50) 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

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [Conditional-DETR-ResNet50 on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/conditional_detr_resnet50) for usage instructions.

## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Model checkpoint: ResNet50
- Input resolution: 480x480
- Number of parameters: 43.6M
- Model size (float): 166 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Conditional-DETR-ResNet50 | ONNX | float | Snapdragon® X Elite | 21.551 ms | 83 - 83 MB | NPU
| Conditional-DETR-ResNet50 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 15.521 ms | 0 - 379 MB | NPU
| Conditional-DETR-ResNet50 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 21.561 ms | 0 - 94 MB | NPU
| Conditional-DETR-ResNet50 | ONNX | float | Qualcomm® QCS9075 | 33.407 ms | 4 - 12 MB | NPU
| Conditional-DETR-ResNet50 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 11.775 ms | 0 - 283 MB | NPU
| Conditional-DETR-ResNet50 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 8.926 ms | 0 - 276 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Snapdragon® X Elite | 23.12 ms | 5 - 5 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 16.723 ms | 3 - 436 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 98.38 ms | 0 - 326 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 22.964 ms | 5 - 7 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® SA8775P | 145.739 ms | 1 - 327 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® QCS9075 | 37.338 ms | 7 - 13 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 46.462 ms | 4 - 372 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® SA7255P | 98.38 ms | 0 - 326 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Qualcomm® SA8295P | 34.708 ms | 0 - 279 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 11.53 ms | 5 - 342 MB | NPU
| Conditional-DETR-ResNet50 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 8.878 ms | 5 - 348 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 16.973 ms | 24 - 495 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 92.649 ms | 0 - 367 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 22.1 ms | 0 - 3 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® SA8775P | 29.851 ms | 0 - 426 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® QCS9075 | 33.136 ms | 0 - 93 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 45.579 ms | 0 - 407 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® SA7255P | 92.649 ms | 0 - 367 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Qualcomm® SA8295P | 34.871 ms | 0 - 316 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 11.897 ms | 0 - 361 MB | NPU
| Conditional-DETR-ResNet50 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 9.205 ms | 0 - 462 MB | NPU

## License
* The license for the original implementation of Conditional-DETR-ResNet50 can be found
  [here](https://github.com/huggingface/transformers/blob/main/LICENSE).

## References
* [Conditional {DETR} for Fast Training Convergence](https://arxiv.org/abs/2108.06152)
* [Source Model Implementation](https://github.com/huggingface/transformers/tree/main/src/transformers/models/conditional_detr)

## 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).