Hugging Face's logo

qualcomm
/
DnCNN

Image-to-Image
PyTorch
real_time
android
Model card Files
xet
 Community
DnCNN
File size: 11,365 Bytes
29af016
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
faeb1e9
 
 
 
 
 
29af016
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
faeb1e9
6267764
faeb1e9
 
 
 
 
 
 
 
6267764
faeb1e9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29af016
 
 
 
 
 
 
 
 
 
 
 
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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
---
library_name: pytorch
license: other
tags:
- real_time
- android
pipeline_tag: image-to-image

---

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

# DnCNN: Optimized for Qualcomm Devices

DnCNN is a 17-layer denoising convolutional neural network that uses residual learning to remove Gaussian noise (sigma=25) from grayscale images. The network predicts the noise residual and subtracts it from the input to produce a clean image.

This is based on the implementation of DnCNN found [here](https://github.com/cszn/KAIR).
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/src/qai_hub_models/models/dncnn) 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 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-onnx-float.zip)
| ONNX | w8a8 | Universal | QAIRT 2.45, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-onnx-w8a8.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-qnn_dlc-float.zip)
| QNN_DLC | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-qnn_dlc-w8a8.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-tflite-float.zip)
| TFLITE | w8a8 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/dncnn/releases/v0.56.0/dncnn-tflite-w8a8.zip)

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


### Option 2: Export with Custom Configurations

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

## Model Details

**Model Type:** Model_use_case.image_editing

**Model Stats:**
- Model checkpoint: dncnn_25
- Input resolution: 256x256
- Number of parameters: 555K
- Model size (float): 2.12 MB
- Model size (w8a8): 581 KB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| DnCNN | ONNX | float | Snapdragon® X2 Elite | 4.042 ms | 213 - 213 MB | NPU
| DnCNN | ONNX | float | Snapdragon® X Elite | 7.137 ms | 180 - 180 MB | NPU
| DnCNN | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.17 ms | 2 - 171 MB | NPU
| DnCNN | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 13.831 ms | 1 - 170 MB | NPU
| DnCNN | ONNX | float | Qualcomm® QCS8550 (Proxy) | 6.94 ms | 1 - 4 MB | NPU
| DnCNN | ONNX | float | Qualcomm® QCS8450 | 13.831 ms | 1 - 170 MB | NPU
| DnCNN | ONNX | float | Snapdragon® 8 Elite Mobile | 4.126 ms | 0 - 143 MB | NPU
| DnCNN | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.133 ms | 0 - 141 MB | NPU
| DnCNN | ONNX | float | Qualcomm® QCS9075 | 14.5 ms | 0 - 46 MB | NPU
| DnCNN | ONNX | float | Qualcomm® QCS8750 | 4.126 ms | 0 - 143 MB | NPU
| DnCNN | ONNX | float | Qualcomm® QCS7181 | 7.137 ms | 180 - 180 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® X2 Elite | 1.042 ms | 213 - 213 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® X Elite | 1.875 ms | 149 - 149 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.337 ms | 0 - 37 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.401 ms | 0 - 44 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS6490 | 7.86 ms | 0 - 46 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS8550 (Proxy) | 1.797 ms | 0 - 23 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS8450 | 2.401 ms | 0 - 44 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.76 ms | 0 - 27 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® 7 Gen 4 Mobile | 3.261 ms | 0 - 143 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCM6690 | 40.034 ms | 0 - 139 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS9075 | 1.946 ms | 0 - 46 MB | NPU
| DnCNN | ONNX | w8a8 | Snapdragon® 8 Elite Mobile | 1.23 ms | 0 - 30 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS7790 | 3.261 ms | 0 - 143 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS8750 | 1.23 ms | 0 - 30 MB | NPU
| DnCNN | ONNX | w8a8 | Qualcomm® QCS7181 | 1.875 ms | 149 - 149 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® X2 Elite | 4.176 ms | 0 - 0 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® X Elite | 7.206 ms | 0 - 0 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 5.05 ms | 0 - 177 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 13.731 ms | 0 - 176 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS8275 | 55.986 ms | 0 - 141 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 6.7 ms | 0 - 2 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS8450 | 13.731 ms | 0 - 176 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 4.035 ms | 0 - 140 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® SA8295P | 15.291 ms | 0 - 139 MB | NPU
| DnCNN | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.974 ms | 0 - 145 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® SA7255P | 55.986 ms | 0 - 141 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS9075 | 14.179 ms | 2 - 4 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS8750 | 4.035 ms | 0 - 140 MB | NPU
| DnCNN | QNN_DLC | float | Qualcomm® QCS7181 | 7.206 ms | 0 - 0 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® X2 Elite | 1.172 ms | 0 - 0 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® X Elite | 2.01 ms | 1 - 1 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.327 ms | 0 - 46 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.414 ms | 0 - 53 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS6490 | 7.566 ms | 0 - 2 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS8275 | 7.608 ms | 0 - 28 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS8550 (Proxy) | 1.792 ms | 0 - 1 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS8450 | 2.414 ms | 0 - 53 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.753 ms | 0 - 28 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® 7 Gen 4 Mobile | 3.255 ms | 0 - 140 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCM6690 | 38.856 ms | 0 - 140 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS9075 | 1.928 ms | 2 - 4 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® SA7255P | 7.608 ms | 0 - 28 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Snapdragon® 8 Elite Mobile | 1.207 ms | 0 - 30 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® SA8295P | 4.206 ms | 0 - 26 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS7790 | 3.255 ms | 0 - 140 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS8750 | 1.207 ms | 0 - 30 MB | NPU
| DnCNN | QNN_DLC | w8a8 | Qualcomm® QCS7181 | 2.01 ms | 1 - 1 MB | NPU
| DnCNN | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 5.199 ms | 0 - 179 MB | NPU
| DnCNN | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 14.1 ms | 1 - 180 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® QCS8275 | 56.398 ms | 0 - 142 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 6.946 ms | 0 - 3 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® SA8775P | 2195.166 ms | 0 - 26 MB | GPU
| DnCNN | TFLITE | float | Qualcomm® SA8650P | 2195.166 ms | 0 - 26 MB | GPU
| DnCNN | TFLITE | float | Qualcomm® SA8255P | 2195.166 ms | 0 - 26 MB | GPU
| DnCNN | TFLITE | float | Qualcomm® QCS8450 | 14.1 ms | 1 - 180 MB | NPU
| DnCNN | TFLITE | float | Snapdragon® 8 Elite Mobile | 4.128 ms | 0 - 143 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® SA8295P | 15.612 ms | 0 - 141 MB | NPU
| DnCNN | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.097 ms | 0 - 147 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® SA7255P | 56.398 ms | 0 - 142 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® QCS9075 | 14.225 ms | 0 - 4 MB | NPU
| DnCNN | TFLITE | float | Qualcomm® QCS8750 | 4.128 ms | 0 - 143 MB | NPU
| DnCNN | TFLITE | w8a8 | Snapdragon® 8 Gen 3 Mobile | 1.302 ms | 0 - 46 MB | NPU
| DnCNN | TFLITE | w8a8 | Snapdragon® 8 Gen 1 Mobile | 2.315 ms | 0 - 51 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS6490 | 7.798 ms | 0 - 3 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS8275 | 7.521 ms | 0 - 29 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS8550 (Proxy) | 1.721 ms | 0 - 3 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® SA8775P | 2203.903 ms | 1 - 26 MB | GPU
| DnCNN | TFLITE | w8a8 | Qualcomm® SA8650P | 2203.903 ms | 1 - 26 MB | GPU
| DnCNN | TFLITE | w8a8 | Qualcomm® SA8255P | 2203.903 ms | 1 - 26 MB | GPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS8450 | 2.315 ms | 0 - 51 MB | NPU
| DnCNN | TFLITE | w8a8 | Snapdragon® 8 Elite Gen 5 Mobile | 0.724 ms | 0 - 29 MB | NPU
| DnCNN | TFLITE | w8a8 | Snapdragon® 7 Gen 4 Mobile | 3.213 ms | 0 - 141 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCM6690 | 38.925 ms | 0 - 140 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS9075 | 1.853 ms | 0 - 3 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® SA7255P | 7.521 ms | 0 - 29 MB | NPU
| DnCNN | TFLITE | w8a8 | Snapdragon® 8 Elite Mobile | 1.175 ms | 0 - 33 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® SA8295P | 4.179 ms | 0 - 27 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS7790 | 3.213 ms | 0 - 141 MB | NPU
| DnCNN | TFLITE | w8a8 | Qualcomm® QCS8750 | 1.175 ms | 0 - 33 MB | NPU

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

## References
* [Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising](https://arxiv.org/abs/1608.03981)
* [Source Model Implementation](https://github.com/cszn/KAIR)

## 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).