Hugging Face's logo

qualcomm
/
SESR-M5

Image-to-Image
PyTorch
android
Model card Files
xet
Community
SESR-M5
File size: 9,458 Bytes 
af3d024
 
 
 
 
abe750a
af3d024
 
 
04f033f
af3d024
3321a96
975a374
af3d024
 
3321a96
 
 
 
 
 
 
 
 
 
 
 
 
 
dd67bab
 
 
 
 
 
3321a96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e32c96b
af3d024
1928792
 
e32c96b
af3d024
 
 
 
 
9cd16c2
af3d024
 
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
 
---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-to-image

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/web-assets/model_demo.png)

# SESR-M5: Optimized for Qualcomm Devices

SESR M5 performs efficient on-device upscaling of images.

This is based on the implementation of SESR-M5 found [here](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/sesr).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/sesr_m5) 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 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.37, ONNX Runtime 1.23.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.42 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.42, TFLite 2.17.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sesr_m5/releases/v0.46.0/sesr_m5-tflite-w8a8.zip)

For more device-specific assets and performance metrics, visit **[SESR-M5 on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/sesr_m5)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/sesr_m5) 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 [SESR-M5 on GitHub](https://github.com/quic/ai-hub-models/blob/main/qai_hub_models/models/sesr_m5) for usage instructions.

## Model Details

**Model Type:** Model_use_case.super_resolution

**Model Stats:**
- Model checkpoint: sesr_m5_3x_checkpoint
- Input resolution: 128x128
- Number of parameters: 343K
- Model size (float): 1.32 MB
- Model size (w8a8): 395 KB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| SESR-M5 | ONNX | float | Snapdragon® X Elite | 2.261 ms | 8 - 8 MB | NPU
| SESR-M5 | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 1.615 ms | 0 - 101 MB | NPU
| SESR-M5 | ONNX | float | Qualcomm® QCS8550 (Proxy) | 2.321 ms | 0 - 4 MB | NPU
| SESR-M5 | ONNX | float | Qualcomm® QCS9075 | 3.402 ms | 7 - 10 MB | NPU
| SESR-M5 | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.315 ms | 0 - 93 MB | NPU
| SESR-M5 | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 1.073 ms | 0 - 92 MB | NPU
| SESR-M5 | ONNX | w8a8 | Snapdragon® X Elite | 0.892 ms | 3 - 3 MB | NPU
| SESR-M5 | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.748 ms | 0 - 100 MB | NPU
| SESR-M5 | ONNX | w8a8 | Qualcomm® QCS6490 | 42.249 ms | 38 - 41 MB | CPU
| SESR-M5 | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.953 ms | 0 - 115 MB | NPU
| SESR-M5 | ONNX | w8a8 | Qualcomm® QCS9075 | 1.222 ms | 0 - 3 MB | NPU
| SESR-M5 | ONNX | w8a8 | Qualcomm® QCM6690 | 30.371 ms | 36 - 43 MB | CPU
| SESR-M5 | ONNX | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.536 ms | 0 - 90 MB | NPU
| SESR-M5 | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 22.496 ms | 38 - 45 MB | CPU
| SESR-M5 | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.488 ms | 0 - 94 MB | NPU
| SESR-M5 | QNN_DLC | float | Snapdragon® X Elite | 2.085 ms | 0 - 0 MB | NPU
| SESR-M5 | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 1.281 ms | 0 - 35 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 10.536 ms | 0 - 24 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 1.911 ms | 0 - 30 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® SA8775P | 3.023 ms | 0 - 25 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® QCS9075 | 3.06 ms | 0 - 5 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 2.788 ms | 0 - 38 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® SA7255P | 10.536 ms | 0 - 24 MB | NPU
| SESR-M5 | QNN_DLC | float | Qualcomm® SA8295P | 3.462 ms | 0 - 21 MB | NPU
| SESR-M5 | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.035 ms | 0 - 28 MB | NPU
| SESR-M5 | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.873 ms | 0 - 27 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Snapdragon® X Elite | 0.756 ms | 0 - 0 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.446 ms | 0 - 31 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 2.729 ms | 2 - 4 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.93 ms | 0 - 24 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.628 ms | 0 - 2 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® SA8775P | 0.843 ms | 0 - 23 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 0.887 ms | 0 - 2 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 6.844 ms | 0 - 134 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.944 ms | 0 - 33 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® SA7255P | 1.93 ms | 0 - 24 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Qualcomm® SA8295P | 1.386 ms | 0 - 19 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.384 ms | 0 - 21 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.84 ms | 0 - 21 MB | NPU
| SESR-M5 | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.316 ms | 0 - 24 MB | NPU
| SESR-M5 | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 1.283 ms | 0 - 35 MB | NPU
| SESR-M5 | TFLITE | float | Qualcomm® QCS8275 (Proxy) | 59.86 ms | 3 - 21 MB | GPU
| SESR-M5 | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 1.872 ms | 0 - 2 MB | NPU
| SESR-M5 | TFLITE | float | Qualcomm® SA8775P | 3.065 ms | 0 - 26 MB | NPU
| SESR-M5 | TFLITE | float | Qualcomm® QCS9075 | 3.086 ms | 3 - 9 MB | NPU
| SESR-M5 | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 2.8 ms | 0 - 37 MB | NPU
| SESR-M5 | TFLITE | float | Qualcomm® SA7255P | 59.86 ms | 3 - 21 MB | GPU
| SESR-M5 | TFLITE | float | Qualcomm® SA8295P | 3.441 ms | 0 - 22 MB | NPU
| SESR-M5 | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 1.041 ms | 0 - 24 MB | NPU
| SESR-M5 | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 0.874 ms | 0 - 28 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 0.448 ms | 0 - 31 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCS6490 | 2.479 ms | 0 - 3 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCS8275 (Proxy) | 1.971 ms | 0 - 23 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 0.641 ms | 0 - 69 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® SA8775P | 0.879 ms | 0 - 24 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCS9075 | 0.91 ms | 0 - 3 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCM6690 | 6.848 ms | 1 - 135 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® QCS8450 (Proxy) | 0.952 ms | 0 - 32 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® SA7255P | 1.971 ms | 0 - 23 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Qualcomm® SA8295P | 1.392 ms | 0 - 20 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Snapdragon® 8 Elite For Galaxy Mobile | 0.383 ms | 0 - 21 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 0.847 ms | 0 - 21 MB | NPU
| SESR-M5 | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.315 ms | 0 - 24 MB | NPU

## License
* The license for the original implementation of SESR-M5 can be found
  [here](https://github.com/quic/aimet-model-zoo/blob/develop/LICENSE.pdf).

## References
* [Collapsible Linear Blocks for Super-Efficient Super Resolution](https://arxiv.org/abs/2103.09404)
* [Source Model Implementation](https://github.com/quic/aimet-model-zoo/tree/develop/aimet_zoo_torch/sesr)

## 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).