OWL-ViT: Optimized for Qualcomm Devices

OWL-ViT (Open-World Localization with Vision Transformers) is an open-vocabulary object detector that uses a CLIP-based ViT-B/32 backbone. Given an image and one or more free-form text queries, the model predicts bounding boxes and confidence scores for each query.

This is based on the implementation of OWL-ViT 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 2.50, ONNX Runtime 1.27.1	Download
ONNX	w8a16	Universal	QAIRT 2.50, ONNX Runtime 1.27.1	Download
QNN_DLC	float	Universal	QAIRT 2.50	Download
QNN_DLC	w8a16	Universal	QAIRT 2.50	Download
TFLITE	float	Universal	QAIRT 2.50	Download

For more device-specific assets and performance metrics, visit OWL-ViT 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 OWL-ViT on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.object_detection

Model Stats:

• Input image resolution: 768x768
• Model checkpoint: google/owlvit-base-patch32
• Model size (float): 613 MB
• Number of parameters: 153M

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
OWL-ViT	ONNX	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	12.21 ms	2 - 479 MB	NPU
OWL-ViT	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	18.217 ms	3 - 485 MB	NPU
OWL-ViT	ONNX	float	Snapdragon® X2 Elite	12.758 ms	8 - 8 MB	NPU
OWL-ViT	ONNX	float	Snapdragon® X Elite	33.658 ms	296 - 296 MB	NPU
OWL-ViT	ONNX	float	Snapdragon® 8 Gen 3 Mobile	23.419 ms	3 - 679 MB	NPU
OWL-ViT	ONNX	float	Snapdragon® 8 Gen 1 Mobile	41.99 ms	7 - 671 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® Dragonwing™ IQ-8275	41.054 ms	7 - 17 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	32.803 ms	0 - 316 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® QCS8450	41.99 ms	7 - 671 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® Dragonwing™ IQ-9075	39.467 ms	7 - 17 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® Dragonwing™ IQ-X7181	33.658 ms	296 - 296 MB	NPU
OWL-ViT	ONNX	float	Qualcomm® Dragonwing™ Q-8750	18.217 ms	3 - 485 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	7.541 ms	1 - 576 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	10.809 ms	1 - 560 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® X2 Elite	8.263 ms	3 - 3 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® X Elite	21.342 ms	174 - 174 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	14.727 ms	4 - 725 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® 8 Gen 1 Mobile	25.699 ms	0 - 726 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ QCS6490	84.949 ms	3 - 9 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-8275	19.894 ms	1 - 8 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	20.74 ms	0 - 192 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® QCS8450	25.699 ms	0 - 726 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-9075	20.865 ms	0 - 7 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ IQ-X7181	21.342 ms	174 - 174 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ Q-7790	34.032 ms	6 - 594 MB	NPU
OWL-ViT	ONNX	w8a16	Qualcomm® Dragonwing™ Q-8750	10.809 ms	1 - 560 MB	NPU
OWL-ViT	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	34.032 ms	6 - 594 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	19.831 ms	0 - 439 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	25.907 ms	7 - 442 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® X2 Elite	18.674 ms	7 - 7 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® X Elite	42.91 ms	7 - 7 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	31.625 ms	7 - 615 MB	NPU
OWL-ViT	QNN_DLC	float	Snapdragon® 8 Gen 1 Mobile	54.368 ms	7 - 589 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-8275	52.057 ms	7 - 16 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	42.246 ms	7 - 9 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® SA8650P	52.042 ms	2 - 483 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® SA8255P	52.042 ms	2 - 483 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® QCS8450	54.368 ms	7 - 589 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-9075	49.439 ms	9 - 18 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® Dragonwing™ IQ-X7181	42.91 ms	7 - 7 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® Dragonwing™ Q-8750	25.907 ms	7 - 442 MB	NPU
OWL-ViT	QNN_DLC	float	Qualcomm® SA8295P	53.597 ms	0 - 407 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	8.735 ms	3 - 520 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	11.485 ms	3 - 504 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Snapdragon® X2 Elite	9.588 ms	3 - 3 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Snapdragon® X Elite	23.188 ms	3 - 3 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	15.673 ms	3 - 662 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-8275	20.866 ms	3 - 10 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ QCS8550 (Proxy)	22.45 ms	4 - 6 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® SA8650P	95.803 ms	0 - 747 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® SA8255P	95.803 ms	0 - 747 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-9075	22.683 ms	3 - 9 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ IQ-X7181	23.188 ms	3 - 3 MB	NPU
OWL-ViT	QNN_DLC	w8a16	Qualcomm® Dragonwing™ Q-8750	11.485 ms	3 - 504 MB	NPU
OWL-ViT	TFLITE	float	Snapdragon® 8 Elite Gen 5 For Galaxy Mobile	11.749 ms	0 - 446 MB	NPU
OWL-ViT	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	17.213 ms	0 - 445 MB	NPU
OWL-ViT	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	22.301 ms	0 - 601 MB	NPU
OWL-ViT	TFLITE	float	Snapdragon® 8 Gen 1 Mobile	39.618 ms	0 - 582 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® Dragonwing™ IQ-8275	39.661 ms	0 - 311 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® Dragonwing™ QCS8550 (Proxy)	31.202 ms	0 - 3 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® SA8650P	39.316 ms	0 - 471 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® SA8255P	39.316 ms	0 - 471 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® QCS8450	39.618 ms	0 - 582 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® Dragonwing™ IQ-9075	38.004 ms	0 - 310 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® Dragonwing™ Q-8750	17.213 ms	0 - 445 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® SA7255P	113.546 ms	0 - 432 MB	NPU
OWL-ViT	TFLITE	float	Qualcomm® SA8295P	41.871 ms	0 - 409 MB	NPU

License

• The license for the original implementation of OWL-ViT can be found here.

References

• Simple Open-Vocabulary Object Detection with Vision Transformers
• 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.