ControlNet-Canny: Optimized for Mobile Deployment

Generating visual arts from text prompt and input guiding image

On-device, high-resolution image synthesis from text and image prompts. ControlNet guides Stable-diffusion with provided input image to generate accurate images from given input prompt.

This model is an implementation of ControlNet-Canny found here.

This repository provides scripts to run ControlNet-Canny on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Model_use_case.image_generation
Model Stats:
- Input: Text prompt and input image as a reference
- Conditioning Input: Canny-Edge
- Text Encoder Number of parameters: 340M
- UNet Number of parameters: 865M
- VAE Decoder Number of parameters: 83M
- ControlNet Number of parameters: 361M
- Model size: 1.4GB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
text_encoder	w8a16	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_CONTEXT_BINARY	5.37 ms	0 - 3 MB	NPU	Use Export Script
text_encoder	w8a16	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	5.903 ms	0 - 10 MB	NPU	Use Export Script
text_encoder	w8a16	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_CONTEXT_BINARY	5.395 ms	0 - 2 MB	NPU	Use Export Script
text_encoder	w8a16	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_CONTEXT_BINARY	5.412 ms	0 - 2 MB	NPU	Use Export Script
text_encoder	w8a16	SA8775P ADP	Qualcomm® SA8775P	QNN_CONTEXT_BINARY	5.903 ms	0 - 10 MB	NPU	Use Export Script
text_encoder	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_CONTEXT_BINARY	5.432 ms	0 - 3 MB	NPU	Use Export Script
text_encoder	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	PRECOMPILED_QNN_ONNX	5.743 ms	0 - 3 MB	NPU	Use Export Script
text_encoder	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	3.872 ms	0 - 18 MB	NPU	Use Export Script
text_encoder	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	4.067 ms	0 - 20 MB	NPU	Use Export Script
text_encoder	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_CONTEXT_BINARY	3.481 ms	0 - 14 MB	NPU	Use Export Script
text_encoder	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	PRECOMPILED_QNN_ONNX	3.255 ms	0 - 13 MB	NPU	Use Export Script
text_encoder	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	5.792 ms	1 - 1 MB	NPU	Use Export Script
text_encoder	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	5.958 ms	158 - 158 MB	NPU	Use Export Script
unet	w8a16	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_CONTEXT_BINARY	110.879 ms	13 - 15 MB	NPU	Use Export Script
unet	w8a16	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	107.956 ms	6 - 13 MB	NPU	Use Export Script
unet	w8a16	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_CONTEXT_BINARY	116.595 ms	13 - 15 MB	NPU	Use Export Script
unet	w8a16	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_CONTEXT_BINARY	115.724 ms	13 - 16 MB	NPU	Use Export Script
unet	w8a16	SA8775P ADP	Qualcomm® SA8775P	QNN_CONTEXT_BINARY	107.956 ms	6 - 13 MB	NPU	Use Export Script
unet	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_CONTEXT_BINARY	117.156 ms	13 - 16 MB	NPU	Use Export Script
unet	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	PRECOMPILED_QNN_ONNX	116.818 ms	0 - 883 MB	NPU	Use Export Script
unet	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	81.085 ms	13 - 31 MB	NPU	Use Export Script
unet	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	84.025 ms	13 - 32 MB	NPU	Use Export Script
unet	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_CONTEXT_BINARY	70.612 ms	13 - 27 MB	NPU	Use Export Script
unet	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	PRECOMPILED_QNN_ONNX	70.807 ms	13 - 28 MB	NPU	Use Export Script
unet	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	116.726 ms	13 - 13 MB	NPU	Use Export Script
unet	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	117.502 ms	829 - 829 MB	NPU	Use Export Script
vae	w8a16	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_CONTEXT_BINARY	268.758 ms	0 - 3 MB	NPU	Use Export Script
vae	w8a16	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	248.983 ms	0 - 10 MB	NPU	Use Export Script
vae	w8a16	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_CONTEXT_BINARY	272.989 ms	0 - 2 MB	NPU	Use Export Script
vae	w8a16	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_CONTEXT_BINARY	284.628 ms	0 - 2 MB	NPU	Use Export Script
vae	w8a16	SA8775P ADP	Qualcomm® SA8775P	QNN_CONTEXT_BINARY	248.983 ms	0 - 10 MB	NPU	Use Export Script
vae	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_CONTEXT_BINARY	270.831 ms	0 - 3 MB	NPU	Use Export Script
vae	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	PRECOMPILED_QNN_ONNX	273.364 ms	0 - 66 MB	NPU	Use Export Script
vae	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	205.993 ms	0 - 18 MB	NPU	Use Export Script
vae	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	204.786 ms	3 - 22 MB	NPU	Use Export Script
vae	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_CONTEXT_BINARY	194.607 ms	0 - 14 MB	NPU	Use Export Script
vae	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	PRECOMPILED_QNN_ONNX	193.998 ms	3 - 17 MB	NPU	Use Export Script
vae	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	266.935 ms	0 - 0 MB	NPU	Use Export Script
vae	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	266.448 ms	63 - 63 MB	NPU	Use Export Script
controlnet	w8a16	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	QNN_CONTEXT_BINARY	83.197 ms	2 - 4 MB	NPU	Use Export Script
controlnet	w8a16	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	81.755 ms	2 - 11 MB	NPU	Use Export Script
controlnet	w8a16	SA8255 (Proxy)	Qualcomm® SA8255P (Proxy)	QNN_CONTEXT_BINARY	83.451 ms	2 - 5 MB	NPU	Use Export Script
controlnet	w8a16	SA8650 (Proxy)	Qualcomm® SA8650P (Proxy)	QNN_CONTEXT_BINARY	83.565 ms	2 - 4 MB	NPU	Use Export Script
controlnet	w8a16	SA8775P ADP	Qualcomm® SA8775P	QNN_CONTEXT_BINARY	81.755 ms	2 - 11 MB	NPU	Use Export Script
controlnet	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	QNN_CONTEXT_BINARY	83.39 ms	2 - 5 MB	NPU	Use Export Script
controlnet	w8a16	Samsung Galaxy S23	Snapdragon® 8 Gen 2 Mobile	PRECOMPILED_QNN_ONNX	86.158 ms	0 - 384 MB	NPU	Use Export Script
controlnet	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	58.723 ms	2 - 21 MB	NPU	Use Export Script
controlnet	w8a16	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	59.623 ms	32 - 50 MB	NPU	Use Export Script
controlnet	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	QNN_CONTEXT_BINARY	56.385 ms	2 - 16 MB	NPU	Use Export Script
controlnet	w8a16	Snapdragon 8 Elite QRD	Snapdragon® 8 Elite Mobile	PRECOMPILED_QNN_ONNX	57.339 ms	31 - 45 MB	NPU	Use Export Script
controlnet	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	85.054 ms	2 - 2 MB	NPU	Use Export Script
controlnet	w8a16	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	80.108 ms	351 - 351 MB	NPU	Use Export Script

Installation

Install the package via pip:

pip install "qai-hub-models[controlnet-canny]"

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.controlnet_canny.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.controlnet_canny.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.controlnet_canny.export

Profiling Results
------------------------------------------------------------
text_encoder
Device                          : cs_8550 (ANDROID 12)                 
Runtime                         : QNN_CONTEXT_BINARY                   
Estimated inference time (ms)   : 5.4                                  
Estimated peak memory usage (MB): [0, 3]                               
Total # Ops                     : 438                                  
Compute Unit(s)                 : npu (438 ops) gpu (0 ops) cpu (0 ops)

------------------------------------------------------------
unet
Device                          : cs_8550 (ANDROID 12)                  
Runtime                         : QNN_CONTEXT_BINARY                    
Estimated inference time (ms)   : 110.9                                 
Estimated peak memory usage (MB): [13, 15]                              
Total # Ops                     : 4055                                  
Compute Unit(s)                 : npu (4055 ops) gpu (0 ops) cpu (0 ops)

------------------------------------------------------------
vae
Device                          : cs_8550 (ANDROID 12)                 
Runtime                         : QNN_CONTEXT_BINARY                   
Estimated inference time (ms)   : 268.8                                
Estimated peak memory usage (MB): [0, 3]                               
Total # Ops                     : 175                                  
Compute Unit(s)                 : npu (175 ops) gpu (0 ops) cpu (0 ops)

------------------------------------------------------------
controlnet
Device                          : cs_8550 (ANDROID 12)                 
Runtime                         : QNN_CONTEXT_BINARY                   
Estimated inference time (ms)   : 83.2                                 
Estimated peak memory usage (MB): [2, 4]                               
Total # Ops                     : 664                                  
Compute Unit(s)                 : npu (664 ops) gpu (0 ops) cpu (0 ops)

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 ControlNet-Canny's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of ControlNet-Canny 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.