Hugging Face's logo

qualcomm
/
SalsaNext

Image Segmentation
PyTorch
bu_auto
android
Model card Files
xet
 Community
SalsaNext
File size: 9,996 Bytes
8922adc
 
 
 
2f02303
8922adc
 
 
 
 
 
 
5842d4a
8922adc
4625f72
8922adc
f09a6d9
5842d4a
 
 
 
 
 
 
 
 
 
 
 
f09a6d9
 
 
 
 
5842d4a
 
 
 
 
 
f09a6d9
5842d4a
 
 
 
 
 
f09a6d9
5842d4a
 
 
 
 
 
 
 
 
 
 
 
 
 
f09a6d9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8922adc
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
---
library_name: pytorch
license: other
tags:
- bu_auto
- android
pipeline_tag: image-segmentation

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/web-assets/model_demo.png)

# SalsaNext: Optimized for Qualcomm Devices

SalsaNext is a LiDAR-based model designed for efficient and accurate semantic segmentation.

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/v0.59.0/src/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).

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.

## 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 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.59.0/salsanext-onnx-float.zip)
| ONNX | w8a16 | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.59.0/salsanext-onnx-w8a16.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.59.0/salsanext-qnn_dlc-float.zip)
| QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.59.0/salsanext-qnn_dlc-w8a16.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/salsanext/releases/v0.59.0/salsanext-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[SalsaNext on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/salsanext)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/salsanext) 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 [SalsaNext on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/salsanext) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: SalsaNext
- Input resolution: 1x5x64x2048
- Number of parameters: 6.71M
- Model size (float): 25.7 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| SalsaNext | ONNX | float | Snapdragon® X2 Elite | 13.861 ms | 35 - 35 MB | NPU
| SalsaNext | ONNX | float | Snapdragon® X Elite | 32.329 ms | 35 - 35 MB | NPU
| SalsaNext | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 25.65 ms | 27 - 359 MB | NPU
| SalsaNext | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 66.054 ms | 3 - 353 MB | NPU
| SalsaNext | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 34.496 ms | 20 - 209 MB | NPU
| SalsaNext | ONNX | float | Qualcomm® QCS8450 | 66.054 ms | 3 - 353 MB | NPU
| SalsaNext | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 40.685 ms | 22 - 25 MB | NPU
| SalsaNext | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 32.329 ms | 35 - 35 MB | NPU
| SalsaNext | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 19.089 ms | 23 - 292 MB | NPU
| SalsaNext | ONNX | float | Snapdragon® 8 Elite Mobile | 19.089 ms | 23 - 292 MB | NPU
| SalsaNext | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.528 ms | 21 - 316 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® X2 Elite | 15.311 ms | 18 - 18 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® X Elite | 33.542 ms | 18 - 18 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 27.244 ms | 13 - 310 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 43.737 ms | 12 - 400 MB | NPU
| SalsaNext | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 36.061 ms | 10 - 19 MB | NPU
| SalsaNext | ONNX | w8a16 | Qualcomm® QCS8450 | 43.737 ms | 12 - 400 MB | NPU
| SalsaNext | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 35.521 ms | 14 - 17 MB | NPU
| SalsaNext | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 33.542 ms | 18 - 18 MB | NPU
| SalsaNext | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 22.78 ms | 14 - 239 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 22.78 ms | 14 - 239 MB | NPU
| SalsaNext | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 17.2 ms | 12 - 245 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® X2 Elite | 13.878 ms | 3 - 3 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® X Elite | 32.065 ms | 3 - 3 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 25.331 ms | 3 - 333 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 65.1 ms | 3 - 346 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 118.795 ms | 1 - 247 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 33.48 ms | 3 - 5 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® SA8775P | 39.305 ms | 1 - 247 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® SA8650P | 39.305 ms | 1 - 247 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® SA8255P | 39.305 ms | 1 - 247 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® QCS8450 | 65.1 ms | 3 - 346 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 39.345 ms | 3 - 17 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 32.065 ms | 3 - 3 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 18.455 ms | 0 - 264 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® SA7255P | 118.795 ms | 1 - 247 MB | NPU
| SalsaNext | QNN_DLC | float | Qualcomm® SA8295P | 51.659 ms | 0 - 260 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 18.455 ms | 0 - 264 MB | NPU
| SalsaNext | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.321 ms | 3 - 293 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 18.54 ms | 1 - 1 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® X Elite | 40.713 ms | 1 - 1 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 33.531 ms | 1 - 365 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Gen 1 Mobile | 70.967 ms | 1 - 419 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8275 | 76.466 ms | 1 - 276 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 40.113 ms | 1 - 3 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® SA8775P | 39.821 ms | 1 - 274 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® SA8650P | 39.821 ms | 1 - 274 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® SA8255P | 39.821 ms | 1 - 274 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® QCS8450 | 70.967 ms | 1 - 419 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 40.23 ms | 1 - 9 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 40.713 ms | 1 - 1 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 27.672 ms | 1 - 297 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® SA7255P | 76.466 ms | 1 - 276 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Qualcomm® SA8295P | 59.105 ms | 1 - 327 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 27.672 ms | 1 - 297 MB | NPU
| SalsaNext | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 19.759 ms | 1 - 348 MB | NPU
| SalsaNext | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 25.421 ms | 0 - 333 MB | NPU
| SalsaNext | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 65.585 ms | 10 - 360 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 118.682 ms | 10 - 257 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 33.293 ms | 10 - 43 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® SA8775P | 39.275 ms | 10 - 257 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® SA8650P | 39.275 ms | 10 - 257 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® SA8255P | 39.275 ms | 10 - 257 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® QCS8450 | 65.585 ms | 10 - 360 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 39.774 ms | 10 - 39 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 18.69 ms | 9 - 272 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® SA7255P | 118.682 ms | 10 - 257 MB | NPU
| SalsaNext | TFLITE | float | Qualcomm® SA8295P | 51.619 ms | 10 - 271 MB | NPU
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite Mobile | 18.69 ms | 9 - 272 MB | NPU
| SalsaNext | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 13.335 ms | 10 - 297 MB | NPU

## License
* The license for the original implementation of SalsaNext can be found
  [here](https://github.com/TiagoCortinhal/SalsaNext/blob/master/LICENSE).



## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).