library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-to-image
QuickSRNetSmall: Optimized for Mobile Deployment
Upscale images and remove image noise
QuickSRNet Small is designed for upscaling images on mobile platforms to sharpen in real-time.
This model is an implementation of QuickSRNetSmall found here.
This repository provides scripts to run QuickSRNetSmall 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:
- Model checkpoint: quicksrnet_small_3x_checkpoint
- Input resolution: 128x128
- Number of parameters: 33.3K
- Model size (float): 133 KB
- Model size (w8a8): 41.7 KB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| QuickSRNetSmall | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 2.319 ms | 0 - 14 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 1.856 ms | 0 - 15 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 1.128 ms | 2 - 31 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 1.024 ms | 0 - 27 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.991 ms | 0 - 7 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.736 ms | 0 - 7 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 1.068 ms | 0 - 9 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 1.405 ms | 0 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 1.075 ms | 0 - 15 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 2.319 ms | 0 - 14 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 1.856 ms | 0 - 15 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.989 ms | 0 - 4 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.74 ms | 2 - 8 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 1.559 ms | 0 - 21 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 1.343 ms | 0 - 24 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.993 ms | 0 - 4 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.754 ms | 0 - 4 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 1.405 ms | 0 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 1.075 ms | 0 - 15 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 0.98 ms | 0 - 6 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 0.734 ms | 0 - 7 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 1.124 ms | 0 - 9 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.607 ms | 0 - 22 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.44 ms | 0 - 27 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 0.726 ms | 0 - 22 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 0.563 ms | 0 - 20 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 0.361 ms | 0 - 26 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 0.644 ms | 0 - 16 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.854 ms | 2 - 2 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 1.11 ms | 8 - 8 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 0.895 ms | 1 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 0.808 ms | 0 - 14 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 0.463 ms | 1 - 29 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 0.453 ms | 0 - 26 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 0.393 ms | 0 - 10 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 0.313 ms | 0 - 4 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 3.227 ms | 0 - 18 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 0.59 ms | 0 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 0.506 ms | 0 - 14 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 0.934 ms | 0 - 18 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN_DLC | 0.87 ms | 0 - 18 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | ONNX | 13.734 ms | 17 - 28 MB | CPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 11.86 ms | 1 - 3 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | ONNX | 12.679 ms | 18 - 20 MB | CPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 0.895 ms | 1 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 0.808 ms | 0 - 14 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 0.389 ms | 0 - 4 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 0.316 ms | 0 - 10 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 0.747 ms | 0 - 24 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 0.692 ms | 0 - 21 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 0.395 ms | 0 - 10 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 0.327 ms | 0 - 10 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 0.59 ms | 0 - 15 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 0.506 ms | 0 - 14 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 0.387 ms | 0 - 3 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 0.336 ms | 0 - 3 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 3.145 ms | 0 - 14 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 0.248 ms | 0 - 24 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 0.216 ms | 0 - 24 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 2.338 ms | 0 - 25 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 0.254 ms | 1 - 21 MB | NPU | QuickSRNetSmall.tflite |
| QuickSRNetSmall | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 0.207 ms | 0 - 24 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 1.977 ms | 0 - 16 MB | NPU | QuickSRNetSmall.onnx.zip |
| QuickSRNetSmall | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 0.433 ms | 6 - 6 MB | NPU | QuickSRNetSmall.dlc |
| QuickSRNetSmall | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 3.277 ms | 31 - 31 MB | NPU | QuickSRNetSmall.onnx.zip |
Installation
Install the package via pip:
pip install qai-hub-models
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.quicksrnetsmall.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.quicksrnetsmall.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.quicksrnetsmall.export
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.quicksrnetsmall 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.quicksrnetsmall.demo --eval-mode 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.quicksrnetsmall.demo -- --eval-mode 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 QuickSRNetSmall's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of QuickSRNetSmall can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- QuickSRNet: Plain Single-Image Super-Resolution Architecture for Faster Inference on Mobile Platforms
- 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.