Efficient and Robust 2D-to-BEV Representation Learning via Geometry-guided Kernel Transformer
Paper • 2206.04584 • Published
GKT: Optimized for Qualcomm Devices
Geometry-guided Kernel Transformer is a machine learning model for generating a birds eye view represenation from the sensors(cameras) mounted on a vehicle.
This is based on the implementation of GKT 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 |
|---|---|---|---|---|
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 1 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50, ONNX Runtime 1.30.0 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-8275 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-9075 | QAIRT 2.50 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | QAIRT 2.50 | Download |
For more device-specific assets and performance metrics, visit GKT 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 GKT on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.driver_assistance
Model Stats:
- Input resolution: 1 x 6 x 3 x 224 x 480
- Model checkpoint: map_segmentation_gkt_7x1_conv_setting2.ckpt
- Model size: 4.66 MB
- Number of parameters: 1.18M
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 59.794 ms | 1 - 10 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 72.547 ms | 2 - 9 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 57.029 ms | 8 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 100.878 ms | 7 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 80.421 ms | 8 - 15 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 1 Mobile | 150.316 ms | 8 - 21 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-8275 | 100.273 ms | 7 - 11 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 110.334 ms | 8 - 12 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8450 | 150.316 ms | 8 - 21 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 106.993 ms | 7 - 11 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 100.878 ms | 7 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 72.547 ms | 2 - 9 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 41.841 ms | 0 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 51.838 ms | 0 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 42.705 ms | 4 - 4 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | 84.026 ms | 5 - 5 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 57.77 ms | 4 - 12 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-8275 | 70.957 ms | 4 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 86.61 ms | 1 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-9075 | 85.407 ms | 4 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-X7181 | 84.026 ms | 5 - 5 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ Q-8750 | 51.838 ms | 0 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 59.803 ms | 7 - 15 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 68.854 ms | 7 - 19 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 56.084 ms | 7 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 100.699 ms | 7 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 81.23 ms | 8 - 16 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 1 Mobile | 152.428 ms | 7 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-8275 | 100.244 ms | 7 - 18 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 106.545 ms | 8 - 10 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 108.502 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8650P | 108.502 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8255P | 108.502 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 | 152.428 ms | 7 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-9075 | 106.971 ms | 7 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ IQ-X7181 | 100.699 ms | 7 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® Dragonwing™ Q-8750 | 68.854 ms | 7 - 19 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 144.822 ms | 0 - 6 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 For Galaxy Mobile | 41.791 ms | 4 - 11 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 51.268 ms | 4 - 12 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 43.04 ms | 4 - 4 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | 84.653 ms | 4 - 4 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 57.082 ms | 4 - 11 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-8275 | 70.829 ms | 3 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 87.023 ms | 4 - 6 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-9075 | 85.262 ms | 3 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ IQ-X7181 | 84.653 ms | 4 - 4 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® Dragonwing™ Q-8750 | 51.268 ms | 4 - 12 MB | NPU |
License
- The license for the original implementation of GKT can be found here.
References
- Efficient and Robust 2D-to-BEV Representation Learning via Geometry-guided Kernel Transformer
- 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.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support