EasyOCR: Optimized for Mobile Deployment
Ready-to-use OCR with 80+ supported languages and all popular writing scripts
EasyOCR is a machine learning model that can recognize text in images. It supports 80+ supported languages and all popular writing scripts.
This model is an implementation of EasyOCR found here.
This repository provides scripts to run EasyOCR on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Image to text
- Model Stats:
- Model checkpoint: easyocr-small-stage1
- Input resolution: 384x384
- Number of parameters (EasyOCRDetector): 20.8M
- Model size (EasyOCRDetector): 79.2 MB
- Number of parameters (EasyOCRRecognizer): 3.84M
- Model size (EasyOCRRecognizer): 14.7 MB
Model | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Precision | Primary Compute Unit | Target Model |
---|---|---|---|---|---|---|---|---|
EasyOCRDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 41.72 ms | 1 - 132 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 37.7 ms | 6 - 17 MB | FP16 | NPU | EasyOCR.so |
EasyOCRDetector | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 41.887 ms | 32 - 121 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 30.07 ms | 14 - 72 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 28.085 ms | 6 - 24 MB | FP16 | NPU | EasyOCR.so |
EasyOCRDetector | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 30.704 ms | 42 - 74 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 29.358 ms | 14 - 48 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 28.786 ms | 6 - 33 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 24.189 ms | 40 - 68 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRDetector | SA7255P ADP | SA7255P | TFLITE | 2113.984 ms | 0 - 30 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | SA7255P ADP | SA7255P | QNN | 2109.673 ms | 1 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 41.306 ms | 9 - 140 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | SA8255 (Proxy) | SA8255P Proxy | QNN | 38.731 ms | 6 - 8 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | SA8295P ADP | SA8295P | TFLITE | 78.453 ms | 16 - 49 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | SA8295P ADP | SA8295P | QNN | 75.057 ms | 0 - 18 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 41.546 ms | 10 - 144 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | SA8650 (Proxy) | SA8650P Proxy | QNN | 38.334 ms | 6 - 7 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | SA8775P ADP | SA8775P | TFLITE | 88.531 ms | 16 - 45 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | SA8775P ADP | SA8775P | QNN | 84.933 ms | 1 - 11 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 2113.984 ms | 0 - 30 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 2109.673 ms | 1 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 38.797 ms | 6 - 8 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 88.531 ms | 16 - 45 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 84.933 ms | 1 - 11 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 82.831 ms | 16 - 77 MB | FP16 | NPU | EasyOCR.tflite |
EasyOCRDetector | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 69.737 ms | 6 - 36 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 38.608 ms | 6 - 6 MB | FP16 | NPU | Use Export Script |
EasyOCRDetector | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 41.643 ms | 66 - 66 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRRecognizer | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | TFLITE | 117.587 ms | 3 - 6 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | QNN | 23.252 ms | 0 - 3 MB | FP16 | NPU | EasyOCR.so |
EasyOCRRecognizer | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 | ONNX | 20.753 ms | 0 - 22 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRRecognizer | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | TFLITE | 109.883 ms | 9 - 29 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | QNN | 16.886 ms | 0 - 18 MB | FP16 | NPU | EasyOCR.so |
EasyOCRRecognizer | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 | ONNX | 16.816 ms | 0 - 25 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRRecognizer | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | TFLITE | 106.439 ms | 20 - 35 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | QNN | 19.025 ms | 0 - 428 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite | ONNX | 14.032 ms | 0 - 22 MB | FP16 | NPU | EasyOCR.onnx |
EasyOCRRecognizer | SA7255P ADP | SA7255P | TFLITE | 571.291 ms | 8 - 18 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | SA7255P ADP | SA7255P | QNN | 281.893 ms | 0 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | SA8255 (Proxy) | SA8255P Proxy | TFLITE | 129.985 ms | 2 - 4 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | SA8255 (Proxy) | SA8255P Proxy | QNN | 23.373 ms | 0 - 2 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | SA8295P ADP | SA8295P | TFLITE | 218.62 ms | 6 - 23 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | SA8295P ADP | SA8295P | QNN | 39.157 ms | 0 - 18 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | SA8650 (Proxy) | SA8650P Proxy | TFLITE | 125.209 ms | 7 - 10 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | SA8650 (Proxy) | SA8650P Proxy | QNN | 23.218 ms | 0 - 3 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | SA8775P ADP | SA8775P | TFLITE | 410.407 ms | 11 - 21 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | SA8775P ADP | SA8775P | QNN | 31.266 ms | 0 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | QCS8275 (Proxy) | QCS8275 Proxy | TFLITE | 571.291 ms | 8 - 18 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | QCS8275 (Proxy) | QCS8275 Proxy | QNN | 281.893 ms | 0 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | QCS8550 (Proxy) | QCS8550 Proxy | TFLITE | 116.297 ms | 0 - 38 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | QCS8550 (Proxy) | QCS8550 Proxy | QNN | 23.278 ms | 0 - 3 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | QCS9075 (Proxy) | QCS9075 Proxy | TFLITE | 410.407 ms | 11 - 21 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | QCS9075 (Proxy) | QCS9075 Proxy | QNN | 31.266 ms | 0 - 10 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | QCS8450 (Proxy) | QCS8450 Proxy | TFLITE | 164.69 ms | 5 - 27 MB | FP32 | CPU | EasyOCR.tflite |
EasyOCRRecognizer | QCS8450 (Proxy) | QCS8450 Proxy | QNN | 36.284 ms | 0 - 170 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 24.48 ms | 0 - 0 MB | FP16 | NPU | Use Export Script |
EasyOCRRecognizer | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 17.61 ms | 0 - 0 MB | FP16 | NPU | EasyOCR.onnx |
Installation
Install the package via pip:
pip install "qai-hub-models[easyocr]"
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.easyocr.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.easyocr.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.easyocr.export
Profiling Results
------------------------------------------------------------
EasyOCRDetector
Device : Samsung Galaxy S23 (13)
Runtime : TFLITE
Estimated inference time (ms) : 41.7
Estimated peak memory usage (MB): [1, 132]
Total # Ops : 42
Compute Unit(s) : NPU (42 ops)
------------------------------------------------------------
EasyOCRRecognizer
Device : Samsung Galaxy S23 (13)
Runtime : TFLITE
Estimated inference time (ms) : 117.6
Estimated peak memory usage (MB): [3, 6]
Total # Ops : 136
Compute Unit(s) : CPU (136 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.easyocr import Model
# Load the model
model = Model.from_pretrained()
detector_model = model.detector
recognizer_model = model.recognizer
# Device
device = hub.Device("Samsung Galaxy S23")
# Trace model
detector_input_shape = detector_model.get_input_spec()
detector_sample_inputs = detector_model.sample_inputs()
traced_detector_model = torch.jit.trace(detector_model, [torch.tensor(data[0]) for _, data in detector_sample_inputs.items()])
# Compile model on a specific device
detector_compile_job = hub.submit_compile_job(
model=traced_detector_model ,
device=device,
input_specs=detector_model.get_input_spec(),
)
# Get target model to run on-device
detector_target_model = detector_compile_job.get_target_model()
# Trace model
recognizer_input_shape = recognizer_model.get_input_spec()
recognizer_sample_inputs = recognizer_model.sample_inputs()
traced_recognizer_model = torch.jit.trace(recognizer_model, [torch.tensor(data[0]) for _, data in recognizer_sample_inputs.items()])
# Compile model on a specific device
recognizer_compile_job = hub.submit_compile_job(
model=traced_recognizer_model ,
device=device,
input_specs=recognizer_model.get_input_spec(),
)
# Get target model to run on-device
recognizer_target_model = recognizer_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.
detector_profile_job = hub.submit_profile_job(
model=detector_target_model,
device=device,
)
recognizer_profile_job = hub.submit_profile_job(
model=recognizer_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.
detector_input_data = detector_model.sample_inputs()
detector_inference_job = hub.submit_inference_job(
model=detector_target_model,
device=device,
inputs=detector_input_data,
)
detector_inference_job.download_output_data()
recognizer_input_data = recognizer_model.sample_inputs()
recognizer_inference_job = hub.submit_inference_job(
model=recognizer_target_model,
device=device,
inputs=recognizer_input_data,
)
recognizer_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 EasyOCR's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of EasyOCR can be found here.
- The license for the compiled assets for on-device deployment can be found here
References
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.