DDColor: Optimized for Qualcomm Devices

DDColor is a coloring algorithm that produces natural, vivid color results from incoming black and white images.

This is based on the implementation of DDColor found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.1	Download
QNN_DLC	float	Universal	QAIRT 2.43	Download
QNN_DLC	w8a8	Universal	QAIRT 2.43	Download
TFLITE	float	Universal	QAIRT 2.43, TFLite 2.17.0	Download
TFLITE	w8a8	Universal	QAIRT 2.43, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit DDColor on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for DDColor on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_editing

Model Stats:

• Model checkpoint: ddcolor_paper_tiny.pth
• Input resolution: 224x224
• Number of parameters: 56.3M
• Model size (float): 215 MB
• Model size (w8a8): 54.8 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
DDColor	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	676.747 ms	1 - 728 MB	NPU
DDColor	QNN_DLC	float	Snapdragon® X2 Elite	713.048 ms	1 - 1 MB	NPU
DDColor	QNN_DLC	float	Snapdragon® X Elite	1146.686 ms	1 - 1 MB	NPU
DDColor	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	831.044 ms	1 - 1368 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	1999.451 ms	1 - 783 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1096.751 ms	1 - 3 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® SA8775P	1108.481 ms	1 - 754 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	1208.928 ms	0 - 485 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® SA7255P	1999.451 ms	1 - 783 MB	NPU
DDColor	QNN_DLC	float	Qualcomm® SA8295P	1257.294 ms	1 - 409 MB	NPU
DDColor	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	838.052 ms	0 - 781 MB	NPU
DDColor	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	936.415 ms	0 - 499 MB	NPU
DDColor	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	1205.208 ms	0 - 567 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	3031.947 ms	0 - 332 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	1733.136 ms	0 - 4 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® SA8775P	1730.83 ms	0 - 331 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® QCM6690	1610.56 ms	104 - 505 MB	CPU
DDColor	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	2116.106 ms	0 - 561 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® SA7255P	3031.947 ms	0 - 332 MB	NPU
DDColor	TFLITE	w8a8	Qualcomm® SA8295P	2043.602 ms	0 - 333 MB	NPU
DDColor	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	967.678 ms	1 - 416 MB	NPU
DDColor	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	492.383 ms	25 - 395 MB	CPU

License

• The license for the original implementation of DDColor can be found here.

References

• DDColor: Towards Photo-Realistic Image Colorization via Dual Decoders
• 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.