v0.48.0
Browse filesSee https://github.com/qualcomm/ai-hub-models/releases/v0.48.0 for changelog.
README.md
CHANGED
|
@@ -14,7 +14,7 @@ pipeline_tag: image-segmentation
|
|
| 14 |
|
| 15 |
SalsaNext is a LiDAR-based model designed for efficient and accurate semantic segmentation.
|
| 16 |
|
| 17 |
-
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/
|
| 18 |
|
| 19 |
Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
|
| 20 |
|
|
@@ -27,25 +27,25 @@ Below are pre-exported model assets ready for deployment.
|
|
| 27 |
|
| 28 |
| Runtime | Precision | Chipset | SDK Versions | Download |
|
| 29 |
|---|---|---|---|---|
|
| 30 |
-
| ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.
|
| 31 |
-
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.
|
| 32 |
-
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.
|
| 33 |
-
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.
|
| 34 |
-
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.
|
| 35 |
|
| 36 |
For more device-specific assets and performance metrics, visit **[SalsaNext on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/salsanext)**.
|
| 37 |
|
| 38 |
|
| 39 |
### Option 2: Export with Custom Configurations
|
| 40 |
|
| 41 |
-
Use the [Qualcomm® AI Hub Models](https://github.com/
|
| 42 |
- Custom weights (e.g., fine-tuned checkpoints)
|
| 43 |
- Custom input shapes
|
| 44 |
- Target device and runtime configurations
|
| 45 |
|
| 46 |
This option is ideal if you need to customize the model beyond the default configuration provided here.
|
| 47 |
|
| 48 |
-
See our repository for [SalsaNext on GitHub](https://github.com/
|
| 49 |
|
| 50 |
## Model Details
|
| 51 |
|
|
@@ -60,48 +60,48 @@ See our repository for [SalsaNext on GitHub](https://github.com/quic/ai-hub-mode
|
|
| 60 |
## Performance Summary
|
| 61 |
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|
| 62 |
|---|---|---|---|---|---|---
|
| 63 |
-
| SalsaNext | ONNX | float | Snapdragon®
|
| 64 |
-
| SalsaNext | ONNX | float | Snapdragon®
|
| 65 |
-
| SalsaNext | ONNX | float |
|
| 66 |
-
| SalsaNext | ONNX | float | Qualcomm®
|
| 67 |
-
| SalsaNext | ONNX | float |
|
| 68 |
-
| SalsaNext | ONNX | float | Snapdragon® 8 Elite
|
| 69 |
-
| SalsaNext | ONNX | float | Snapdragon®
|
| 70 |
-
| SalsaNext | ONNX | w8a16 | Snapdragon®
|
| 71 |
-
| SalsaNext | ONNX | w8a16 | Snapdragon®
|
| 72 |
-
| SalsaNext | ONNX | w8a16 |
|
| 73 |
-
| SalsaNext | ONNX | w8a16 | Qualcomm®
|
| 74 |
-
| SalsaNext | ONNX | w8a16 | Qualcomm®
|
| 75 |
-
| SalsaNext | ONNX | w8a16 | Qualcomm®
|
| 76 |
-
| SalsaNext | ONNX | w8a16 |
|
| 77 |
-
| SalsaNext | ONNX | w8a16 | Snapdragon®
|
| 78 |
-
| SalsaNext | ONNX | w8a16 | Snapdragon®
|
| 79 |
-
| SalsaNext | ONNX | w8a16 | Snapdragon®
|
| 80 |
-
| SalsaNext | QNN_DLC | float | Snapdragon®
|
| 81 |
-
| SalsaNext | QNN_DLC | float | Snapdragon®
|
| 82 |
-
| SalsaNext | QNN_DLC | float |
|
| 83 |
-
| SalsaNext | QNN_DLC | float | Qualcomm®
|
| 84 |
-
| SalsaNext | QNN_DLC | float | Qualcomm®
|
| 85 |
-
| SalsaNext | QNN_DLC | float | Qualcomm®
|
| 86 |
-
| SalsaNext | QNN_DLC | float |
|
| 87 |
-
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite
|
| 88 |
-
| SalsaNext | QNN_DLC | float | Snapdragon®
|
| 89 |
-
| SalsaNext | QNN_DLC | w8a16 | Snapdragon®
|
| 90 |
-
| SalsaNext | QNN_DLC | w8a16 | Snapdragon®
|
| 91 |
-
| SalsaNext | QNN_DLC | w8a16 |
|
| 92 |
-
| SalsaNext | QNN_DLC | w8a16 | Qualcomm®
|
| 93 |
-
| SalsaNext | QNN_DLC | w8a16 | Qualcomm®
|
| 94 |
-
| SalsaNext | QNN_DLC | w8a16 | Qualcomm®
|
| 95 |
-
| SalsaNext | QNN_DLC | w8a16 |
|
| 96 |
-
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite
|
| 97 |
-
| SalsaNext | QNN_DLC | w8a16 | Snapdragon®
|
| 98 |
-
| SalsaNext | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 22.
|
| 99 |
-
| SalsaNext | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 116.
|
| 100 |
-
| SalsaNext | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 30.
|
| 101 |
-
| SalsaNext | TFLITE | float | Qualcomm® QCS9075 | 39.
|
| 102 |
-
| SalsaNext | TFLITE | float | Qualcomm® QCS8450 (Proxy) |
|
| 103 |
-
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.
|
| 104 |
-
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.
|
| 105 |
|
| 106 |
## License
|
| 107 |
* The license for the original implementation of SalsaNext can be found
|
|
|
|
| 14 |
|
| 15 |
SalsaNext is a LiDAR-based model designed for efficient and accurate semantic segmentation.
|
| 16 |
|
| 17 |
+
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/salsanext) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
|
| 18 |
|
| 19 |
Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
|
| 20 |
|
|
|
|
| 27 |
|
| 28 |
| Runtime | Precision | Chipset | SDK Versions | Download |
|
| 29 |
|---|---|---|---|---|
|
| 30 |
+
| ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.48.0/salsanext-onnx-float.zip)
|
| 31 |
+
| ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.24.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.48.0/salsanext-onnx-w8a16.zip)
|
| 32 |
+
| QNN_DLC | float | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.48.0/salsanext-qnn_dlc-float.zip)
|
| 33 |
+
| QNN_DLC | w8a16 | Universal | QAIRT 2.43 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.48.0/salsanext-qnn_dlc-w8a16.zip)
|
| 34 |
+
| TFLITE | float | Universal | QAIRT 2.43, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.48.0/salsanext-tflite-float.zip)
|
| 35 |
|
| 36 |
For more device-specific assets and performance metrics, visit **[SalsaNext on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/salsanext)**.
|
| 37 |
|
| 38 |
|
| 39 |
### Option 2: Export with Custom Configurations
|
| 40 |
|
| 41 |
+
Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/salsanext) Python library to compile and export the model with your own:
|
| 42 |
- Custom weights (e.g., fine-tuned checkpoints)
|
| 43 |
- Custom input shapes
|
| 44 |
- Target device and runtime configurations
|
| 45 |
|
| 46 |
This option is ideal if you need to customize the model beyond the default configuration provided here.
|
| 47 |
|
| 48 |
+
See our repository for [SalsaNext on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/qai_hub_models/models/salsanext) for usage instructions.
|
| 49 |
|
| 50 |
## Model Details
|
| 51 |
|
|
|
|
| 60 |
## Performance Summary
|
| 61 |
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|
| 62 |
|---|---|---|---|---|---|---
|
| 63 |
+
| SalsaNext | ONNX | float | Snapdragon® X2 Elite | 15.016 ms | 33 - 33 MB | NPU
|
| 64 |
+
| SalsaNext | ONNX | float | Snapdragon® X Elite | 32.478 ms | 33 - 33 MB | NPU
|
| 65 |
+
| SalsaNext | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 25.581 ms | 24 - 293 MB | NPU
|
| 66 |
+
| SalsaNext | ONNX | float | Qualcomm® QCS8550 (Proxy) | 34.553 ms | 23 - 28 MB | NPU
|
| 67 |
+
| SalsaNext | ONNX | float | Qualcomm® QCS9075 | 39.948 ms | 23 - 25 MB | NPU
|
| 68 |
+
| SalsaNext | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.594 ms | 21 - 209 MB | NPU
|
| 69 |
+
| SalsaNext | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.584 ms | 24 - 215 MB | NPU
|
| 70 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® X2 Elite | 16.662 ms | 16 - 16 MB | NPU
|
| 71 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® X Elite | 34.291 ms | 18 - 18 MB | NPU
|
| 72 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 29.256 ms | 13 - 297 MB | NPU
|
| 73 |
+
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS6490 | 3780.547 ms | 357 - 361 MB | CPU
|
| 74 |
+
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 36.503 ms | 10 - 371 MB | NPU
|
| 75 |
+
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS9075 | 34.256 ms | 12 - 15 MB | NPU
|
| 76 |
+
| SalsaNext | ONNX | w8a16 | Qualcomm® QCM6690 | 2030.785 ms | 290 - 299 MB | CPU
|
| 77 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 23.989 ms | 14 - 216 MB | NPU
|
| 78 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 1831.046 ms | 368 - 377 MB | CPU
|
| 79 |
+
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 18.524 ms | 12 - 208 MB | NPU
|
| 80 |
+
| SalsaNext | QNN_DLC | float | Snapdragon® X2 Elite | 15.188 ms | 3 - 3 MB | NPU
|
| 81 |
+
| SalsaNext | QNN_DLC | float | Snapdragon® X Elite | 30.098 ms | 3 - 3 MB | NPU
|
| 82 |
+
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 22.614 ms | 3 - 264 MB | NPU
|
| 83 |
+
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 116.643 ms | 2 - 178 MB | NPU
|
| 84 |
+
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 30.492 ms | 3 - 4 MB | NPU
|
| 85 |
+
| SalsaNext | QNN_DLC | float | Qualcomm® QCS9075 | 40.037 ms | 3 - 17 MB | NPU
|
| 86 |
+
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 54.964 ms | 2 - 297 MB | NPU
|
| 87 |
+
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 20.92 ms | 0 - 185 MB | NPU
|
| 88 |
+
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.375 ms | 3 - 190 MB | NPU
|
| 89 |
+
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 18.338 ms | 1 - 1 MB | NPU
|
| 90 |
+
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X Elite | 43.785 ms | 1 - 1 MB | NPU
|
| 91 |
+
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 34.222 ms | 0 - 366 MB | NPU
|
| 92 |
+
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8275 (Proxy) | 76.982 ms | 1 - 257 MB | NPU
|
| 93 |
+
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 45.857 ms | 1 - 3 MB | NPU
|
| 94 |
+
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 50.235 ms | 1 - 9 MB | NPU
|
| 95 |
+
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8450 (Proxy) | 82.618 ms | 1 - 424 MB | NPU
|
| 96 |
+
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 28.685 ms | 1 - 276 MB | NPU
|
| 97 |
+
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 19.382 ms | 1 - 314 MB | NPU
|
| 98 |
+
| SalsaNext | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 22.77 ms | 9 - 276 MB | NPU
|
| 99 |
+
| SalsaNext | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 116.827 ms | 0 - 178 MB | NPU
|
| 100 |
+
| SalsaNext | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 30.691 ms | 6 - 8 MB | NPU
|
| 101 |
+
| SalsaNext | TFLITE | float | Qualcomm® QCS9075 | 39.693 ms | 10 - 38 MB | NPU
|
| 102 |
+
| SalsaNext | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 54.381 ms | 10 - 310 MB | NPU
|
| 103 |
+
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 18.446 ms | 10 - 197 MB | NPU
|
| 104 |
+
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.544 ms | 10 - 199 MB | NPU
|
| 105 |
|
| 106 |
## License
|
| 107 |
* The license for the original implementation of SalsaNext can be found
|