---
library_name: pytorch
license: apache-2.0
tags:
- real_time
- android
pipeline_tag: image-segmentation

---

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

# Segformer-Base: Optimized for Qualcomm Devices

Segformer Base is a machine learning model that predicts masks and classes of objects in an image.

This is based on the implementation of Segformer-Base found [here](https://github.com/huggingface/transformers/tree/main/src/transformers/models/segformer).
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/v0.64.0/src/qai_hub_models/models/segformer_base) 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 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-onnx-w8a16.zip)
| ONNX | w8a8 | Universal | QAIRT 2.50, ONNX Runtime 1.30.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-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/segformer_base/releases/v0.64.0/segformer_base-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-qnn_dlc-w8a16.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-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/segformer_base/releases/v0.64.0/segformer_base-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.50 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/releases/v0.64.0/segformer_base-tflite-w8a8.zip)

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


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.64.0/src/qai_hub_models/models/segformer_base) 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 [Segformer-Base on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.64.0/src/qai_hub_models/models/segformer_base) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Input resolution: 512x512
- Model checkpoint: nvidia/segformer-b0-finetuned-ade-512-512
- Model size (float): 14.4 MB
- Model size (w8a16): 4.57 MB
- Model size (w8a8): 3.90 MB
- Number of output classes: 150
- Number of parameters: 3.75M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| Segformer-Base | ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 13.07 ms | 21 - 567 MB | NPU
| Segformer-Base | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 14.769 ms | 21 - 562 MB | NPU
| Segformer-Base | ONNX | float | Snapdragon® X2 Elite | 13.576 ms | 35 - 35 MB | NPU
| Segformer-Base | ONNX | float | Snapdragon® X Elite | 26.023 ms | 35 - 35 MB | NPU
| Segformer-Base | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 17.846 ms | 23 - 631 MB | NPU
| Segformer-Base | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 27.312 ms | 13 - 619 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 27.293 ms | 22 - 26 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 25.19 ms | 19 - 28 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® QCS8450 | 27.312 ms | 13 - 619 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 30.133 ms | 23 - 27 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 26.023 ms | 35 - 35 MB | NPU
| Segformer-Base | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 14.769 ms | 21 - 562 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 7.357 ms | 12 - 211 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 8.656 ms | 13 - 203 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® X2 Elite | 6.68 ms | 18 - 18 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® X Elite | 15.863 ms | 17 - 17 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 11.163 ms | 12 - 239 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 18.956 ms | 12 - 237 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 60.557 ms | 14 - 17 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 14.901 ms | 11 - 15 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 15.883 ms | 9 - 15 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® QCS8450 | 18.956 ms | 12 - 237 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 16.461 ms | 14 - 18 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 15.863 ms | 17 - 17 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 169.358 ms | 3 - 230 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 24.442 ms | 12 - 221 MB | NPU
| Segformer-Base | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 8.656 ms | 13 - 203 MB | NPU
| Segformer-Base | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 24.442 ms | 12 - 221 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 1.972 ms | 0 - 191 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.433 ms | 0 - 182 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® X2 Elite | 2.098 ms | 9 - 9 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® X Elite | 5.066 ms | 5 - 5 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.296 ms | 0 - 209 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 5.843 ms | 6 - 218 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 17.078 ms | 9 - 12 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 4.717 ms | 5 - 10 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 4.827 ms | 5 - 12 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® QCS8450 | 5.843 ms | 6 - 218 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 5.256 ms | 7 - 10 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 5.066 ms | 5 - 5 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 32.247 ms | 0 - 197 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 5.92 ms | 8 - 201 MB | NPU
| Segformer-Base | ONNX | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 2.433 ms | 0 - 182 MB | NPU
| Segformer-Base | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 5.92 ms | 8 - 201 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 11.626 ms | 3 - 187 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 13.051 ms | 0 - 176 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® X2 Elite | 12.22 ms | 3 - 3 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® X Elite | 23.652 ms | 3 - 3 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 16.35 ms | 0 - 211 MB | NPU
| Segformer-Base | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 26.857 ms | 3 - 214 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-8275 | 23.383 ms | 3 - 18 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 22.619 ms | 3 - 5 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® SA8650P | 24.435 ms | 0 - 193 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® SA8255P | 24.435 ms | 0 - 193 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® QCS8450 | 26.857 ms | 3 - 214 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 25.489 ms | 3 - 17 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 23.652 ms | 3 - 3 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 13.051 ms | 0 - 176 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® SA7255P | 49.156 ms | 0 - 172 MB | NPU
| Segformer-Base | QNN_DLC | float | Qualcomm® SA8295P | 30.32 ms | 0 - 171 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 5.689 ms | 2 - 189 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 7.562 ms | 2 - 185 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 6.394 ms | 2 - 2 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® X Elite | 15.203 ms | 2 - 2 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 10.516 ms | 2 - 213 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-8275 | 14.411 ms | 2 - 11 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 14.658 ms | 2 - 3 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® SA8650P | 20.163 ms | 2 - 206 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® SA8255P | 20.163 ms | 2 - 206 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 15.462 ms | 1 - 9 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 15.203 ms | 2 - 2 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 101.75 ms | 0 - 213 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 17.315 ms | 2 - 198 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 7.562 ms | 2 - 185 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Qualcomm® SA7255P | 26.172 ms | 2 - 180 MB | NPU
| Segformer-Base | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 17.315 ms | 2 - 198 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 2.246 ms | 1 - 179 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 2.823 ms | 1 - 170 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 2.606 ms | 1 - 1 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® X Elite | 6.421 ms | 1 - 1 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 3.964 ms | 0 - 191 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 7.072 ms | 0 - 195 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 17.181 ms | 2 - 7 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 5.53 ms | 0 - 6 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.788 ms | 1 - 3 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® SA8775P | 6.413 ms | 1 - 172 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® SA8650P | 6.413 ms | 1 - 172 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® SA8255P | 6.413 ms | 1 - 172 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 7.072 ms | 0 - 195 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 6.196 ms | 3 - 8 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ IQ-X7181 | 6.421 ms | 1 - 1 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 27.81 ms | 1 - 183 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 6.288 ms | 1 - 175 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 2.823 ms | 1 - 170 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® SA7255P | 10.847 ms | 1 - 170 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Qualcomm® SA8295P | 7.627 ms | 1 - 175 MB | NPU
| Segformer-Base | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 6.288 ms | 1 - 175 MB | NPU
| Segformer-Base | TFLITE | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 13.728 ms | 8 - 193 MB | NPU
| Segformer-Base | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 15.17 ms | 9 - 195 MB | NPU
| Segformer-Base | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 18.017 ms | 8 - 225 MB | NPU
| Segformer-Base | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 31.811 ms | 9 - 227 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-8275 | 25.248 ms | 9 - 33 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 25.193 ms | 9 - 14 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® SA8650P | 24.464 ms | 9 - 202 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® SA8255P | 24.464 ms | 9 - 202 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® QCS8450 | 31.811 ms | 9 - 227 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 27.519 ms | 8 - 31 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 15.17 ms | 9 - 195 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® SA7255P | 52.079 ms | 10 - 187 MB | NPU
| Segformer-Base | TFLITE | float | Qualcomm® SA8295P | 30.787 ms | 9 - 190 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 4.64 ms | 2 - 188 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 5.454 ms | 2 - 195 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 6.706 ms | 2 - 202 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 10.313 ms | 2 - 213 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS6490 | 128.061 ms | 13 - 49 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-8275 | 9.111 ms | 2 - 13 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 9.305 ms | 2 - 11 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® SA8775P | 10.058 ms | 2 - 188 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® SA8650P | 10.058 ms | 2 - 188 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® SA8255P | 10.058 ms | 2 - 188 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® QCS8450 | 10.313 ms | 2 - 213 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ IQ-9075 | 9.748 ms | 2 - 13 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-6690 | 98.048 ms | 13 - 187 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-7790 | 39.687 ms | 15 - 184 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® Dragonwing™ Q-8750 | 5.454 ms | 2 - 195 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Qualcomm® SA8295P | 11.843 ms | 2 - 191 MB | NPU
| Segformer-Base | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 39.687 ms | 15 - 184 MB | NPU

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

## References
* [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203)
* [Source Model Implementation](https://github.com/huggingface/transformers/tree/main/src/transformers/models/segformer)

## 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).