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/resnet34_ssd1200/web-assets/model_demo.png)

# ResNet34-SSD: Optimized for Qualcomm Devices

ResNet34-SSD is a single-stage object detection model that integrates the ResNet34 backbone with the SSD (Single Shot MultiBox Detector) framework. It is optimized for real-time detection tasks and supports multiple deployment backends including PyTorch, TensorFlow, and ONNX.

This is based on the implementation of ResNet34-SSD found [here](https://github.com/mlcommons/inference/tree/33894a19c4af6207f7cfdda75f84570f04836de5/vision/classification_and_detection).
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/main/src/qai_hub_models/models/resnet34_ssd1200) 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 2.45, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet34_ssd1200/releases/v0.56.0/resnet34_ssd1200-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet34_ssd1200/releases/v0.56.0/resnet34_ssd1200-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/resnet34_ssd1200/releases/v0.56.0/resnet34_ssd1200-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[ResNet34-SSD on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/resnet34_ssd1200)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/resnet34_ssd1200) 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 [ResNet34-SSD on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/resnet34_ssd1200) for usage instructions.

## Model Details

**Model Type:** Model_use_case.object_detection

**Model Stats:**
- Model checkpoint: resnet34-ssd1200
- Input resolution: 1x3x1200x1200
- Number of parameters: 20.0M
- Model size (float): 76.2 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| ResNet34-SSD | ONNX | float | Snapdragon® X2 Elite | 43.349 ms | 164 - 164 MB | NPU
| ResNet34-SSD | ONNX | float | Snapdragon® X Elite | 88.337 ms | 132 - 132 MB | NPU
| ResNet34-SSD | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 63.557 ms | 17 - 507 MB | NPU
| ResNet34-SSD | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 177.876 ms | 17 - 437 MB | NPU
| ResNet34-SSD | ONNX | float | Qualcomm® QCS8550 (Proxy) | 86.745 ms | 0 - 32 MB | NPU
| ResNet34-SSD | ONNX | float | Qualcomm® QCS8450 | 177.876 ms | 17 - 437 MB | NPU
| ResNet34-SSD | ONNX | float | Snapdragon® 8 Elite Mobile | 52.014 ms | 1 - 422 MB | NPU
| ResNet34-SSD | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 38.567 ms | 1 - 494 MB | NPU
| ResNet34-SSD | ONNX | float | Qualcomm® QCS9075 | 153.507 ms | 16 - 78 MB | NPU
| ResNet34-SSD | ONNX | float | Qualcomm® QCS8750 | 52.014 ms | 1 - 422 MB | NPU
| ResNet34-SSD | ONNX | float | Qualcomm® QCS7181 | 88.337 ms | 132 - 132 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® X2 Elite | 62.483 ms | 17 - 17 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® X Elite | 129.833 ms | 17 - 17 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 85.129 ms | 16 - 605 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 259.163 ms | 16 - 522 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS8275 | 481.99 ms | 16 - 383 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 134.915 ms | 17 - 19 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS8450 | 259.163 ms | 16 - 522 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 67.26 ms | 16 - 391 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® SA8295P | 183.207 ms | 1 - 329 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 51.973 ms | 0 - 547 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® SA7255P | 481.99 ms | 16 - 383 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS9075 | 194.463 ms | 17 - 35 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS8750 | 67.26 ms | 16 - 391 MB | NPU
| ResNet34-SSD | QNN_DLC | float | Qualcomm® QCS7181 | 129.833 ms | 17 - 17 MB | NPU
| ResNet34-SSD | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 107.277 ms | 0 - 542 MB | NPU
| ResNet34-SSD | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 243.208 ms | 1 - 619 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® QCS8275 | 513.425 ms | 0 - 378 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 143.989 ms | 6 - 9 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® SA8775P | 462.59 ms | 18 - 107 MB | CPU
| ResNet34-SSD | TFLITE | float | Qualcomm® SA8650P | 462.59 ms | 18 - 107 MB | CPU
| ResNet34-SSD | TFLITE | float | Qualcomm® SA8255P | 462.59 ms | 18 - 107 MB | CPU
| ResNet34-SSD | TFLITE | float | Qualcomm® QCS8450 | 243.208 ms | 1 - 619 MB | NPU
| ResNet34-SSD | TFLITE | float | Snapdragon® 8 Elite Mobile | 87.467 ms | 0 - 403 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® SA8295P | 201.906 ms | 0 - 353 MB | NPU
| ResNet34-SSD | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 71.678 ms | 0 - 570 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® SA7255P | 513.425 ms | 0 - 378 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® QCS9075 | 199.528 ms | 0 - 64 MB | NPU
| ResNet34-SSD | TFLITE | float | Qualcomm® QCS8750 | 87.467 ms | 0 - 403 MB | NPU

## License
* The license for the original implementation of ResNet34-SSD can be found
  [here](https://github.com/mlcommons/inference/blob/33894a19c4af6207f7cfdda75f84570f04836de5/LICENSE.md).

## References
* [Source Model Implementation](https://github.com/mlcommons/inference/tree/33894a19c4af6207f7cfdda75f84570f04836de5/vision/classification_and_detection)

## 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).