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](https://github.com/hustvl/GKT/ https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf). 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 |
|---|---|---|---|---|
| QNN_CONTEXT_BINARY | float | qualcomm_qcs8450_proxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_qcs8550_proxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_qcs9075 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_sa7255p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_sa8295p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_sa8775p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8_elite_for_galaxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8_elite_gen5 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_8gen3 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_x2_elite | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | float | qualcomm_snapdragon_x_elite | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_qcs8450_proxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_qcs8550_proxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_qcs9075 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_sa7255p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_sa8295p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_sa8775p | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_snapdragon_8_elite_for_galaxy | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_snapdragon_8_elite_gen5 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_snapdragon_8gen3 | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_snapdragon_x2_elite | QAIRT 2.43 | Download |
| QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | qualcomm_snapdragon_x_elite | QAIRT 2.43 | 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:
- Model checkpoint: map_segmentation_gkt_7x1_conv_setting2.ckpt
- Input resolution: 1 x 6 x 3 x 224 x 480
- Number of parameters: 1.18M
- Model size: 4.66 MB
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 Mobile | 56.763 ms | 8 - 18 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X2 Elite | 61.693 ms | 6 - 6 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® X Elite | 105.184 ms | 7 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Gen 3 Mobile | 79.746 ms | 8 - 15 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS8550 (Proxy) | 110.33 ms | 8 - 11 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Qualcomm® QCS9075 | 110.359 ms | 7 - 10 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.426 ms | 1 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 72.082 ms | 4 - 14 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 81.256 ms | 7 - 7 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | 130.804 ms | 6 - 6 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 98.087 ms | 4 - 11 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 131.116 ms | 0 - 8 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 133.72 ms | 4 - 6 MB | NPU |
| GKT | PRECOMPILED_QNN_ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 84.691 ms | 0 - 12 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite Gen 5 Mobile | 58.545 ms | 8 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X2 Elite | 61.811 ms | 7 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® X Elite | 104.743 ms | 7 - 7 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Gen 3 Mobile | 78.541 ms | 8 - 15 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8275 (Proxy) | 183.394 ms | 0 - 10 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8550 (Proxy) | 109.092 ms | 8 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8775P | 110.662 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS9075 | 110.064 ms | 7 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® QCS8450 (Proxy) | 209.237 ms | 8 - 17 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA7255P | 183.394 ms | 0 - 10 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Qualcomm® SA8295P | 140.529 ms | 1 - 6 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | float | Snapdragon® 8 Elite For Galaxy Mobile | 71.275 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 38.541 ms | 4 - 14 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 41.741 ms | 4 - 4 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® X Elite | 93.081 ms | 4 - 4 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 64.835 ms | 4 - 11 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8275 (Proxy) | 141.361 ms | 0 - 8 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 95.303 ms | 5 - 6 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8775P | 93.425 ms | 0 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 93.84 ms | 4 - 9 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® QCS8450 (Proxy) | 145.368 ms | 4 - 13 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA7255P | 141.361 ms | 0 - 8 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Qualcomm® SA8295P | 109.583 ms | 0 - 6 MB | NPU |
| GKT | QNN_CONTEXT_BINARY | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 49.154 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](https://github.com/hustvl/GKT/ https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
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