library_name: pytorch
license: other
tags:
- android
pipeline_tag: other
CVT: Optimized for Qualcomm Devices
Cross-View Transformer generates real-time bird's-eye view maps from multiple vehicle cameras for autonomous driving.
This is based on the implementation of CVT 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 1.24.3 | Download |
| ONNX | w8a16_mixed_fp16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.3 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| QNN_DLC | w8a16_mixed_fp16 | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit CVT 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 CVT on GitHub for usage instructions.
Model Details
Model Type: Model_use_case.driver_assistance
Model Stats:
- Model checkpoint: vehicles_50k.pt
- Inference latency: RealTime
- Input resolution: 1x6x3x224x480
- Number of parameters: 1.33M
- Model size (float): 5.18 MB
Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |
|---|---|---|---|---|---|---|
| CVT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 183.433 ms | 8 - 2258 MB | NPU |
| CVT | ONNX | float | Snapdragon® X2 Elite | 183.052 ms | 20 - 20 MB | NPU |
| CVT | ONNX | float | Snapdragon® X Elite | 282.825 ms | 18 - 18 MB | NPU |
| CVT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 224.699 ms | 8 - 2578 MB | NPU |
| CVT | ONNX | float | Qualcomm® QCS8550 (Proxy) | 318.335 ms | 0 - 24 MB | NPU |
| CVT | ONNX | float | Qualcomm® QCS9075 | 326.55 ms | 7 - 10 MB | NPU |
| CVT | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 188.812 ms | 3 - 2222 MB | NPU |
| CVT | ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 216.427 ms | 6 - 1317 MB | NPU |
| CVT | ONNX | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 221.839 ms | 18 - 18 MB | NPU |
| CVT | ONNX | w8a16_mixed_fp16 | Snapdragon® X Elite | 355.386 ms | 15 - 15 MB | NPU |
| CVT | ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 303.476 ms | 6 - 1626 MB | NPU |
| CVT | ONNX | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 253.463 ms | 4 - 1117 MB | NPU |
| CVT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 179.073 ms | 7 - 2308 MB | NPU |
| CVT | QNN_DLC | float | Snapdragon® X2 Elite | 181.432 ms | 7 - 7 MB | NPU |
| CVT | QNN_DLC | float | Snapdragon® X Elite | 270.434 ms | 7 - 7 MB | NPU |
| CVT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 227.639 ms | 7 - 2588 MB | NPU |
| CVT | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 328.998 ms | 0 - 1092 MB | NPU |
| CVT | QNN_DLC | float | Qualcomm® QCS9075 | 324.86 ms | 7 - 17 MB | NPU |
| CVT | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 439.412 ms | 6 - 2717 MB | NPU |
| CVT | QNN_DLC | float | Qualcomm® SA8295P | 336.32 ms | 0 - 2330 MB | NPU |
| CVT | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 185.53 ms | 7 - 2383 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Snapdragon® 8 Elite Gen 5 Mobile | 203.486 ms | 4 - 2395 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Snapdragon® X2 Elite | 206.333 ms | 4 - 4 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Snapdragon® X Elite | 320.411 ms | 4 - 4 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Snapdragon® 8 Gen 3 Mobile | 245.379 ms | 4 - 2556 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Qualcomm® QCS8275 (Proxy) | 472.502 ms | 2 - 2284 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Qualcomm® QCS8550 (Proxy) | 357.202 ms | 4 - 1500 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Qualcomm® SA8775P | 323.593 ms | 0 - 2282 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Qualcomm® QCS9075 | 395.971 ms | 5 - 11 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Qualcomm® SA7255P | 472.502 ms | 2 - 2284 MB | NPU |
| CVT | QNN_DLC | w8a16_mixed_fp16 | Snapdragon® 8 Elite For Galaxy Mobile | 211.53 ms | 4 - 2347 MB | NPU |
| CVT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 174.276 ms | 0 - 2332 MB | NPU |
| CVT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 233.495 ms | 0 - 2667 MB | NPU |
| CVT | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 434.684 ms | 0 - 2303 MB | NPU |
| CVT | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 324.164 ms | 0 - 4 MB | NPU |
| CVT | TFLITE | float | Qualcomm® SA8775P | 302.238 ms | 0 - 2303 MB | NPU |
| CVT | TFLITE | float | Qualcomm® QCS9075 | 331.09 ms | 0 - 36 MB | NPU |
| CVT | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 443.407 ms | 0 - 2740 MB | NPU |
| CVT | TFLITE | float | Qualcomm® SA7255P | 434.684 ms | 0 - 2303 MB | NPU |
| CVT | TFLITE | float | Qualcomm® SA8295P | 346.35 ms | 0 - 2304 MB | NPU |
| CVT | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 193.625 ms | 0 - 2398 MB | NPU |
License
- The license for the original implementation of CVT 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.