Real-ESRGAN-x4plus: Optimized for Mobile Deployment
Upscale images and remove image noise
Real-ESRGAN is a machine learning model that upscales an image with minimal loss in quality. The implementation is a derivative of the Real-ESRGAN-x4plus architecture, a larger and more powerful version compared to the Real-ESRGAN-general-x4v3 architecture.
This model is an implementation of Real-ESRGAN-x4plus found here.
This repository provides scripts to run Real-ESRGAN-x4plus on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Model_use_case.super_resolution
- Model Stats:
- Number of parameters: 16.7M
- Model size: 67.1 MB
- Model checkpoint: RealESRGAN_x4plus
- Input resolution: 128x128
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| Real-ESRGAN-x4plus | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 458.491 ms | 2 - 179 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 448.624 ms | 0 - 10 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 122.711 ms | 3 - 162 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 142.778 ms | 0 - 137 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 72.198 ms | 0 - 54 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 69.552 ms | 0 - 3 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 111.216 ms | 3 - 179 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 105.355 ms | 0 - 14 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 458.491 ms | 2 - 179 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 448.624 ms | 0 - 10 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 69.583 ms | 0 - 54 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 62.691 ms | 0 - 3 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 114.026 ms | 0 - 150 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 110.924 ms | 0 - 17 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 73.602 ms | 3 - 40 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 63.036 ms | 0 - 10 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 111.216 ms | 3 - 179 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 105.355 ms | 0 - 14 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 73.776 ms | 3 - 40 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 65.241 ms | 0 - 48 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 70.097 ms | 0 - 105 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 54.046 ms | 3 - 185 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 49.815 ms | 0 - 125 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 50.79 ms | 6 - 199 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 47.024 ms | 3 - 182 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 39.839 ms | 0 - 117 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 42.923 ms | 0 - 121 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 64.733 ms | 0 - 0 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 65.618 ms | 37 - 37 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 74.301 ms | 1 - 186 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 71.536 ms | 0 - 10 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 38.102 ms | 1 - 193 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 40.751 ms | 0 - 187 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 25.126 ms | 0 - 35 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 25.663 ms | 0 - 3 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 23.914 ms | 1 - 185 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 23.8 ms | 0 - 14 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 107.45 ms | 1 - 267 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 167.186 ms | 0 - 12 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 1758.33 ms | 0 - 71 MB | GPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 74.301 ms | 1 - 186 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN | 71.536 ms | 0 - 10 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 25.145 ms | 0 - 35 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 25.565 ms | 0 - 2 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 43.011 ms | 1 - 186 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN | 37.219 ms | 0 - 18 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 25.4 ms | 0 - 35 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 25.688 ms | 0 - 2 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 23.914 ms | 1 - 185 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN | 23.8 ms | 0 - 14 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 25.248 ms | 0 - 36 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 25.668 ms | 0 - 40 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 26.771 ms | 8 - 69 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 17.866 ms | 1 - 188 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 17.712 ms | 0 - 185 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 19.208 ms | 8 - 250 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 16.908 ms | 1 - 186 MB | NPU | Real-ESRGAN-x4plus.tflite |
| Real-ESRGAN-x4plus | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 15.747 ms | 0 - 187 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 18.804 ms | 8 - 239 MB | NPU | Real-ESRGAN-x4plus.onnx |
| Real-ESRGAN-x4plus | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 26.641 ms | 0 - 0 MB | NPU | Use Export Script |
| Real-ESRGAN-x4plus | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 34.095 ms | 22 - 22 MB | NPU | Real-ESRGAN-x4plus.onnx |
Installation
Install the package via pip:
pip install "qai-hub-models[real-esrgan-x4plus]"
Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
Sign-in to Qualcomm® AI Hub with your
Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.
With this API token, you can configure your client to run models on the cloud hosted devices.
qai-hub configure --api_token API_TOKEN
Navigate to docs for more information.
Demo off target
The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.
python -m qai_hub_models.models.real_esrgan_x4plus.demo
The above demo runs a reference implementation of pre-processing, model inference, and post processing.
NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).
%run -m qai_hub_models.models.real_esrgan_x4plus.demo
Run model on a cloud-hosted device
In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:
- Performance check on-device on a cloud-hosted device
- Downloads compiled assets that can be deployed on-device for Android.
- Accuracy check between PyTorch and on-device outputs.
python -m qai_hub_models.models.real_esrgan_x4plus.export
Profiling Results
------------------------------------------------------------
Real-ESRGAN-x4plus
Device : cs_8275 (ANDROID 14)
Runtime : TFLITE
Estimated inference time (ms) : 458.5
Estimated peak memory usage (MB): [2, 179]
Total # Ops : 1025
Compute Unit(s) : npu (1025 ops) gpu (0 ops) cpu (0 ops)
How does this work?
This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:
Step 1: Compile model for on-device deployment
To compile a PyTorch model for on-device deployment, we first trace the model
in memory using the jit.trace and then call the submit_compile_job API.
import torch
import qai_hub as hub
from qai_hub_models.models.real_esrgan_x4plus import Model
# Load the model
torch_model = Model.from_pretrained()
# Device
device = hub.Device("Samsung Galaxy S24")
# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()
pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
# Compile model on a specific device
compile_job = hub.submit_compile_job(
model=pt_model,
device=device,
input_specs=torch_model.get_input_spec(),
)
# Get target model to run on-device
target_model = compile_job.get_target_model()
Step 2: Performance profiling on cloud-hosted device
After compiling models from step 1. Models can be profiled model on-device using the
target_model. Note that this scripts runs the model on a device automatically
provisioned in the cloud. Once the job is submitted, you can navigate to a
provided job URL to view a variety of on-device performance metrics.
profile_job = hub.submit_profile_job(
model=target_model,
device=device,
)
Step 3: Verify on-device accuracy
To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.
input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
model=target_model,
device=device,
inputs=input_data,
)
on_device_output = inference_job.download_output_data()
With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.
Note: This on-device profiling and inference requires access to Qualcomm® AI Hub. Sign up for access.
Run demo on a cloud-hosted device
You can also run the demo on-device.
python -m qai_hub_models.models.real_esrgan_x4plus.demo --on-device
NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).
%run -m qai_hub_models.models.real_esrgan_x4plus.demo -- --on-device
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tfliteexport): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.soexport ): This sample app provides instructions on how to use the.soshared library in an Android application.
View on Qualcomm® AI Hub
Get more details on Real-ESRGAN-x4plus's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of Real-ESRGAN-x4plus can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data
- Source Model Implementation
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.