README.md

---
library_name: pytorch
license: other
tags:
- backbone
- bu_auto
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/web-assets/model_demo.png)

# DenseNet-121: Optimized for Qualcomm Devices

Densenet is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of DenseNet-121 found [here](https://github.com/pytorch/vision/blob/main/torchvision/models/densenet.py).
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/v0.60.0/src/qai_hub_models/models/densenet121) 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 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-onnx-w8a16.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/densenet121/releases/v0.60.0/densenet121-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[DenseNet-121 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/densenet121)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/densenet121) 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 [DenseNet-121 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.60.0/src/qai_hub_models/models/densenet121) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Input resolution: 224x224
- Model checkpoint: Imagenet
- Model size (float): 30.5 MB
- Model size (w8a16): 8.72 MB
- Model size (w8a8): 8.30 MB
- Number of parameters: 7.99M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| DenseNet-121 | ONNX | float | Snapdragon® X2 Elite | 0.73 ms | 2 - 2 MB | NPU
| DenseNet-121 | ONNX | float | Snapdragon® X Elite | 1.586 ms | 16 - 16 MB | NPU
| DenseNet-121 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.058 ms | 0 - 93 MB | NPU
| DenseNet-121 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 2.621 ms | 1 - 92 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 2.463 ms | 1 - 4 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.526 ms | 0 - 46 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® QCS8450 | 2.621 ms | 1 - 92 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 2.382 ms | 1 - 3 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 1.586 ms | 16 - 16 MB | NPU
| DenseNet-121 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 0.782 ms | 0 - 44 MB | NPU
| DenseNet-121 | ONNX | float | Snapdragon® 8 Elite Mobile | 0.782 ms | 0 - 44 MB | NPU
| DenseNet-121 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.668 ms | 0 - 47 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® X2 Elite | 0.845 ms | 1 - 1 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® X Elite | 1.824 ms | 9 - 9 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.261 ms | 0 - 105 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 2.238 ms | 0 - 106 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 7.205 ms | 0 - 3 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 1.858 ms | 0 - 4 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.791 ms | 0 - 120 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® QCS8450 | 2.238 ms | 0 - 106 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 1.981 ms | 0 - 3 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 1.824 ms | 9 - 9 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 10.628 ms | 0 - 201 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 1.901 ms | 0 - 196 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 0.891 ms | 0 - 84 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 0.891 ms | 0 - 84 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 0.782 ms | 0 - 83 MB | NPU
| DenseNet-121 | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1.901 ms | 0 - 196 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.485 ms | 1 - 1 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® X Elite | 0.9 ms | 8 - 8 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.608 ms | 0 - 110 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.505 ms | 0 - 105 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 4.856 ms | 0 - 3 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 0.929 ms | 0 - 4 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 0.846 ms | 0 - 12 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® QCS8450 | 1.505 ms | 0 - 105 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 0.981 ms | 0 - 3 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 0.9 ms | 8 - 8 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 7.027 ms | 0 - 205 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 1.016 ms | 0 - 86 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.513 ms | 0 - 87 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 0.513 ms | 0 - 87 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.458 ms | 0 - 90 MB | NPU
| DenseNet-121 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 1.016 ms | 0 - 86 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® X2 Elite | 0.955 ms | 1 - 1 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® X Elite | 2.011 ms | 1 - 1 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.216 ms | 0 - 71 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 3.286 ms | 0 - 76 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 2.517 ms | 1 - 4 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 7.878 ms | 1 - 37 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.795 ms | 1 - 2 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® SA8775P | 2.699 ms | 1 - 40 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® SA8650P | 2.699 ms | 1 - 40 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® SA8255P | 2.699 ms | 1 - 40 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® QCS8450 | 3.286 ms | 0 - 76 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 2.532 ms | 3 - 5 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 2.011 ms | 1 - 1 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 0.882 ms | 0 - 42 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® SA7255P | 7.878 ms | 1 - 37 MB | NPU
| DenseNet-121 | QNN_DLC | float | Qualcomm® SA8295P | 3.08 ms | 0 - 35 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 0.882 ms | 0 - 42 MB | NPU
| DenseNet-121 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.72 ms | 1 - 40 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 1.438 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® X Elite | 3.122 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 1.985 ms | 0 - 85 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® 8 Gen 1 Mobile | 4.427 ms | 0 - 87 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 2.742 ms | 0 - 3 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 5.795 ms | 0 - 65 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.806 ms | 0 - 31 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® SA8775P | 3.198 ms | 0 - 68 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® SA8650P | 3.198 ms | 0 - 68 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® SA8255P | 3.198 ms | 0 - 68 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® QCS8450 | 4.427 ms | 0 - 87 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 3.016 ms | 0 - 2 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 3.122 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 1.713 ms | 0 - 58 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® SA7255P | 5.795 ms | 0 - 65 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Qualcomm® SA8295P | 4.935 ms | 0 - 63 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 1.713 ms | 0 - 58 MB | NPU
| DenseNet-121 | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 1.105 ms | 0 - 76 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 1.139 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® X Elite | 2.628 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.68 ms | 0 - 81 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.902 ms | 0 - 81 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 2.336 ms | 0 - 3 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 4.981 ms | 0 - 60 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 2.354 ms | 0 - 2 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 2.675 ms | 0 - 61 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® SA8650P | 2.675 ms | 0 - 61 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® SA8255P | 2.675 ms | 0 - 61 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 2.902 ms | 0 - 81 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 2.514 ms | 0 - 2 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 2.628 ms | 0 - 0 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 1.044 ms | 0 - 60 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 4.981 ms | 0 - 60 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 3.371 ms | 0 - 58 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 1.044 ms | 0 - 60 MB | NPU
| DenseNet-121 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.9 ms | 0 - 68 MB | NPU
| DenseNet-121 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.206 ms | 0 - 92 MB | NPU
| DenseNet-121 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 3.287 ms | 0 - 90 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 2.525 ms | 0 - 19 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 7.911 ms | 0 - 52 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.817 ms | 0 - 10 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® SA8775P | 2.713 ms | 0 - 56 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® SA8650P | 2.713 ms | 0 - 56 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® SA8255P | 2.713 ms | 0 - 56 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® QCS8450 | 3.287 ms | 0 - 90 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 2.547 ms | 0 - 18 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 0.882 ms | 0 - 50 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® SA7255P | 7.911 ms | 0 - 52 MB | NPU
| DenseNet-121 | TFLITE | float | Qualcomm® SA8295P | 3.085 ms | 0 - 49 MB | NPU
| DenseNet-121 | TFLITE | float | Snapdragon® 8 Elite Mobile | 0.882 ms | 0 - 50 MB | NPU
| DenseNet-121 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.723 ms | 0 - 55 MB | NPU

## License
* The license for the original implementation of DenseNet-121 can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [Densely Connected Convolutional Networks](https://arxiv.org/abs/1608.06993)
* [Source Model Implementation](https://github.com/pytorch/vision/blob/main/torchvision/models/densenet.py)

## 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).