README.md

---
library_name: pytorch
license: apache-2.0
tags:
- backbone
- android
pipeline_tag: image-classification

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/web-assets/model_demo.png)

# Inception-v4: Optimized for Qualcomm Devices

InceptionNetV4 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 Inception-v4 found [here](https://huggingface.co/timm/inception_v4.tf_in1k).
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/v0.63.0/src/qai_hub_models/models/inception_v4) 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 2.50, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/inception_v4/releases/v0.63.0/inception_v4-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[Inception-v4 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/inception_v4)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.63.0/src/qai_hub_models/models/inception_v4) 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 [Inception-v4 on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.63.0/src/qai_hub_models/models/inception_v4) for usage instructions.

## Model Details

**Model Type:** Model_use_case.image_classification

**Model Stats:**
- Input resolution: 299x299
- Model checkpoint: Imagenet
- Model size (float): 163 MB
- Model size (w8a8): 41.6 MB
- Number of parameters: 42.6M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Inception-v4 | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.923 ms | 0 - 56 MB | NPU
| Inception-v4 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.356 ms | 0 - 54 MB | NPU
| Inception-v4 | ONNX | float | Snapdragon® X2 Elite | 2.125 ms | 1 - 1 MB | NPU
| Inception-v4 | ONNX | float | Snapdragon® X Elite | 3.96 ms | 82 - 82 MB | NPU
| Inception-v4 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 2.961 ms | 0 - 104 MB | NPU
| Inception-v4 | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 6.767 ms | 1 - 108 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 8.109 ms | 1 - 6 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.989 ms | 0 - 97 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® QCS8450 | 6.767 ms | 1 - 108 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 7.02 ms | 1 - 5 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.96 ms | 82 - 82 MB | NPU
| Inception-v4 | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 2.356 ms | 0 - 54 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.699 ms | 0 - 97 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.839 ms | 0 - 96 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® X2 Elite | 0.665 ms | 1 - 1 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® X Elite | 1.302 ms | 42 - 42 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.96 ms | 0 - 140 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.719 ms | 0 - 142 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 6.14 ms | 0 - 4 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 1.517 ms | 0 - 5 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.32 ms | 0 - 50 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® QCS8450 | 1.719 ms | 0 - 142 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.478 ms | 0 - 4 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.302 ms | 42 - 42 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 26.365 ms | 0 - 212 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 2.321 ms | 0 - 105 MB | NPU
| Inception-v4 | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.839 ms | 0 - 96 MB | NPU
| Inception-v4 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 2.321 ms | 0 - 105 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.839 ms | 1 - 55 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.289 ms | 0 - 47 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® X2 Elite | 2.261 ms | 1 - 1 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® X Elite | 3.956 ms | 1 - 1 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 2.834 ms | 0 - 101 MB | NPU
| Inception-v4 | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 6.651 ms | 1 - 102 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 7.85 ms | 1 - 5 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.806 ms | 1 - 3 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® SA8775P | 7.056 ms | 1 - 49 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® SA8650P | 7.056 ms | 1 - 49 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® SA8255P | 7.056 ms | 1 - 49 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® QCS8450 | 6.651 ms | 1 - 102 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 6.786 ms | 3 - 6 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 3.956 ms | 1 - 1 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 2.289 ms | 0 - 47 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® SA7255P | 27.19 ms | 1 - 47 MB | NPU
| Inception-v4 | QNN_DLC | float | Qualcomm® SA8295P | 7.574 ms | 0 - 49 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.698 ms | 0 - 84 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.856 ms | 0 - 83 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 0.833 ms | 0 - 0 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® X Elite | 1.403 ms | 0 - 0 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.955 ms | 0 - 117 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.7 ms | 0 - 122 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 6.167 ms | 2 - 5 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 1.478 ms | 0 - 3 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.317 ms | 0 - 2 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 1.7 ms | 0 - 122 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.407 ms | 2 - 5 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 1.403 ms | 0 - 0 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 22.521 ms | 0 - 200 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 2.292 ms | 0 - 89 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.856 ms | 0 - 83 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 2.473 ms | 0 - 81 MB | NPU
| Inception-v4 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 2.292 ms | 0 - 89 MB | NPU
| Inception-v4 | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.836 ms | 0 - 112 MB | NPU
| Inception-v4 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 2.287 ms | 0 - 112 MB | NPU
| Inception-v4 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 2.812 ms | 0 - 209 MB | NPU
| Inception-v4 | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 6.575 ms | 0 - 213 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 7.86 ms | 0 - 86 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 3.802 ms | 0 - 2 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® SA8775P | 7.076 ms | 0 - 111 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® SA8650P | 7.076 ms | 0 - 111 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® SA8255P | 7.076 ms | 0 - 111 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® QCS8450 | 6.575 ms | 0 - 213 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 6.799 ms | 0 - 85 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 2.287 ms | 0 - 112 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® SA7255P | 27.234 ms | 0 - 109 MB | NPU
| Inception-v4 | TFLITE | float | Qualcomm® SA8295P | 7.588 ms | 0 - 113 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 0.694 ms | 0 - 81 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.837 ms | 0 - 85 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.956 ms | 0 - 115 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 1.68 ms | 0 - 122 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 5.842 ms | 0 - 44 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 1.488 ms | 0 - 45 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 1.297 ms | 0 - 3 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® SA8775P | 1.581 ms | 0 - 82 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® SA8650P | 1.581 ms | 0 - 82 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® SA8255P | 1.581 ms | 0 - 82 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® QCS8450 | 1.68 ms | 0 - 122 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 1.434 ms | 0 - 44 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 22.165 ms | 0 - 198 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 2.295 ms | 0 - 85 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 0.837 ms | 0 - 85 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® SA7255P | 149.112 ms | 1 - 32 MB | GPU
| Inception-v4 | TFLITE | w8a8 | Qualcomm® SA8295P | 2.454 ms | 0 - 78 MB | NPU
| Inception-v4 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 2.295 ms | 0 - 85 MB | NPU

## License
* The license for the original implementation of Inception-v4 can be found
  [here](https://github.com/huggingface/pytorch-image-models/tree/main?tab=Apache-2.0-1-ov-file).

## References
* [Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning](https://arxiv.org/abs/1602.07261)
* [Source Model Implementation](https://huggingface.co/timm/inception_v4.tf_in1k)

## 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).