HRNet-W48-OCR: Optimized for Mobile Deployment
Semantic segmentation in higher resolution
HRNet-W48-OCR is a machine learning model that can segment images from the Cityscape dataset. It has lightweight and hardware-efficient operations and thus delivers significant speedup on diverse hardware platforms
This model is an implementation of HRNet-W48-OCR found here.
This repository provides scripts to run HRNet-W48-OCR on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Model_use_case.semantic_segmentation
- Model Stats:
- Model checkpoint: hrnet_ocr_cs_8162_torch11.pth
- Input resolution: 2048x1024
- Number of output classes: 19
- Number of parameters: 70.3M
- Model size (float): 268 MB
- Model size (w8a16): 70.3 MB
Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
---|---|---|---|---|---|---|---|---|
HRNet-W48-OCR | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 132698.221 ms | 11 - 991 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 135327.545 ms | 4 - 986 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 74365.943 ms | 7 - 1518 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 75510.086 ms | 41 - 1131 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 81392.046 ms | 4 - 84 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 78352.436 ms | 27 - 112 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 76371.06 ms | 3 - 983 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 77693.764 ms | 1 - 984 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 80866.958 ms | 4 - 85 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 77970.107 ms | 24 - 114 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 96206.757 ms | 2 - 321 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 55819.115 ms | 2 - 1110 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 56870.122 ms | 30 - 971 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 65554.29 ms | 31 - 703 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 43713.273 ms | 0 - 959 MB | NPU | HRNet-W48-OCR.tflite |
HRNet-W48-OCR | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 47296.252 ms | 24 - 1012 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 50076.088 ms | 30 - 755 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 81056.503 ms | 324 - 324 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 81404.252 ms | 131 - 131 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 12186.267 ms | 462 - 1229 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 9037.36 ms | 12 - 754 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 8700.597 ms | 12 - 131 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 7686.179 ms | 5 - 771 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 8647.697 ms | 12 - 132 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 8054.785 ms | 8 - 256 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 6643.812 ms | 12 - 724 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 6265.718 ms | 14 - 967 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 10385.889 ms | 14 - 868 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 10273.66 ms | 14 - 1224 MB | NPU | HRNet-W48-OCR.onnx |
HRNet-W48-OCR | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 8524.584 ms | 238 - 238 MB | NPU | HRNet-W48-OCR.dlc |
HRNet-W48-OCR | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 8644.149 ms | 88 - 88 MB | NPU | HRNet-W48-OCR.onnx |
Installation
Install the package via pip:
pip install "qai-hub-models[hrnet-w48-ocr]"
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.hrnet_w48_ocr.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.hrnet_w48_ocr.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.hrnet_w48_ocr.export
Profiling Results
------------------------------------------------------------
HRNet-W48-OCR
Device : cs_8275 (ANDROID 14)
Runtime : TFLITE
Estimated inference time (ms) : 132698.2
Estimated peak memory usage (MB): [11, 991]
Total # Ops : 579
Compute Unit(s) : npu (579 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.hrnet_w48_ocr 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.
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tflite
export): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.so
export ): This sample app provides instructions on how to use the.so
shared library in an Android application.
View on Qualcomm® AI Hub
Get more details on HRNet-W48-OCR's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of HRNet-W48-OCR can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
- Segmentation Transformer: Object-Contextual Representations for Semantic Segmentation
- 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.
- Downloads last month
- 25