Hugging Face's logo

qualcomm
/
NAFNet-DeNoise

Image-to-Image
PyTorch
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
491645a
·
verified ·
1 Parent(s): c1453a0

v0.55.0

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.55.0 for changelog.

Files changed (3) hide show
  1. LICENSE +1 -0
  2. README.md +141 -0
  3. release_assets.json +46 -0
LICENSE ADDED
@@ -0,0 +1 @@
 
 
1
+ The license of the original trained model can be found at https://github.com/megvii-research/NAFNet/blob/main/LICENSE.
README.md ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - android
6
+ pipeline_tag: image-to-image
7
+
8
+ ---
9
+
10
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/web-assets/model_demo.png)
11
+
12
+ # NAFNet-DeNoise: Optimized for Qualcomm Devices
13
+
14
+ NAFNET is designed for lightweight real-time denoising of images.
15
+
16
+ This is based on the implementation of NAFNet-DeNoise found [here](https://github.com/megvii-research/NAFNet.git).
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/src/qai_hub_models/models/nafnet_denoise) 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
+
21
+ ## Getting Started
22
+ There are two ways to deploy this model on your device:
23
+
24
+ ### Option 1: Download Pre-Exported Models
25
+
26
+ 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 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-onnx-float.zip)
31
+ | ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-onnx-w8a16.zip)
32
+ | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-qnn_dlc-float.zip)
33
+ | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-qnn_dlc-w8a16.zip)
34
+ | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-tflite-float.zip)
35
+
36
+ For more device-specific assets and performance metrics, visit **[NAFNet-DeNoise on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/nafnet_denoise)**.
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/src/qai_hub_models/models/nafnet_denoise) 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 [NAFNet-DeNoise on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/nafnet_denoise) for usage instructions.
49
+
50
+ ## Model Details
51
+
52
+ **Model Type:** Model_use_case.image_editing
53
+
54
+ **Model Stats:**
55
+ - Model checkpoint: NAFNet-SIDD-width64
56
+ - Input resolution: 256x256
57
+ - Number of parameters: 115.98M
58
+ - Model size (float): 463.93 MB
59
+
60
+ ## Performance Summary
61
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
62
+ |---|---|---|---|---|---|---
63
+ | NAFNet-DeNoise | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 15.267 ms | 2 - 568 MB | NPU
64
+ | NAFNet-DeNoise | ONNX | float | Snapdragon® X2 Elite | 15.346 ms | 180 - 180 MB | NPU
65
+ | NAFNet-DeNoise | ONNX | float | Snapdragon® X Elite | 33.92 ms | 227 - 227 MB | NPU
66
+ | NAFNet-DeNoise | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 27.542 ms | 2 - 879 MB | NPU
67
+ | NAFNet-DeNoise | ONNX | float | Qualcomm® QCS8550 (Proxy) | 34.799 ms | 0 - 273 MB | NPU
68
+ | NAFNet-DeNoise | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 20.561 ms | 2 - 508 MB | NPU
69
+ | NAFNet-DeNoise | ONNX | float | Qualcomm® QCS9075 | 46.169 ms | 2 - 47 MB | NPU
70
+ | NAFNet-DeNoise | ONNX | float | Qualcomm® QCS8750 | 20.561 ms | 2 - 508 MB | NPU
71
+ | NAFNet-DeNoise | ONNX | float | Qualcomm® QCS7181 | 33.92 ms | 227 - 227 MB | NPU
72
+ | NAFNet-DeNoise | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 15.899 ms | 1 - 715 MB | NPU
73
+ | NAFNet-DeNoise | ONNX | w8a16 | Snapdragon® X Elite | 40.406 ms | 148 - 148 MB | NPU
74
+ | NAFNet-DeNoise | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 28.566 ms | 1 - 737 MB | NPU
75
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS6490 | 4731.65 ms | 397 - 425 MB | CPU
76
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 39.492 ms | 0 - 501 MB | NPU
77
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCM6690 | 2302.284 ms | 397 - 420 MB | CPU
78
+ | NAFNet-DeNoise | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 2033.157 ms | 375 - 396 MB | CPU
79
+ | NAFNet-DeNoise | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 22.928 ms | 1 - 583 MB | NPU
80
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS9075 | 51.312 ms | 1 - 46 MB | NPU
81
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS7790 | 2033.157 ms | 375 - 396 MB | CPU
82
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS8750 | 22.928 ms | 1 - 583 MB | NPU
83
+ | NAFNet-DeNoise | ONNX | w8a16 | Qualcomm® QCS7181 | 40.406 ms | 148 - 148 MB | NPU
84
+ | NAFNet-DeNoise | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.647 ms | 1 - 614 MB | NPU
85
+ | NAFNet-DeNoise | QNN_DLC | float | Snapdragon® X2 Elite | 15.716 ms | 1 - 1 MB | NPU
86
+ | NAFNet-DeNoise | QNN_DLC | float | Snapdragon® X Elite | 35.805 ms | 1 - 1 MB | NPU
87
+ | NAFNet-DeNoise | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 28.197 ms | 0 - 888 MB | NPU
88
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS8275 | 136.377 ms | 1 - 545 MB | NPU
89
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 37.385 ms | 1 - 564 MB | NPU
90
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® SA8775P | 43.49 ms | 1 - 558 MB | NPU
91
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® SA8650P | 43.49 ms | 1 - 558 MB | NPU
92
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® SA8255P | 43.49 ms | 1 - 558 MB | NPU
93
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 61.64 ms | 1 - 729 MB | NPU
94
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® SA7255P | 136.377 ms | 1 - 545 MB | NPU
95
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® SA8295P | 47.971 ms | 1 - 399 MB | NPU
96
+ | NAFNet-DeNoise | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 20.0 ms | 1 - 543 MB | NPU
97
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS9075 | 46.091 ms | 1 - 4 MB | NPU
98
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS8750 | 20.0 ms | 1 - 543 MB | NPU
99
+ | NAFNet-DeNoise | QNN_DLC | float | Qualcomm® QCS7181 | 35.805 ms | 1 - 1 MB | NPU
100
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 13.139 ms | 0 - 699 MB | NPU
101
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 15.128 ms | 0 - 0 MB | NPU
102
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® X Elite | 33.213 ms | 0 - 0 MB | NPU
103
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 22.718 ms | 0 - 688 MB | NPU
104
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 65.079 ms | 1 - 577 MB | NPU
105
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 31.918 ms | 0 - 211 MB | NPU
106
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® SA8775P | 31.626 ms | 1 - 577 MB | NPU
107
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® SA8650P | 31.626 ms | 1 - 577 MB | NPU
108
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® SA8255P | 31.626 ms | 1 - 577 MB | NPU
109
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 367.326 ms | 0 - 953 MB | NPU
110
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® SA7255P | 65.079 ms | 1 - 577 MB | NPU
111
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 39.568 ms | 0 - 816 MB | NPU
112
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 20.863 ms | 0 - 584 MB | NPU
113
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 45.973 ms | 0 - 3 MB | NPU
114
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS7790 | 39.568 ms | 0 - 816 MB | NPU
115
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS8750 | 20.863 ms | 0 - 584 MB | NPU
116
+ | NAFNet-DeNoise | QNN_DLC | w8a16 | Qualcomm® QCS7181 | 33.213 ms | 0 - 0 MB | NPU
117
+ | NAFNet-DeNoise | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 14.822 ms | 1 - 841 MB | NPU
118
+ | NAFNet-DeNoise | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 28.805 ms | 1 - 1128 MB | NPU
119
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® QCS8275 | 136.992 ms | 1 - 765 MB | NPU
120
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 36.384 ms | 1 - 4 MB | NPU
121
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® SA8775P | 43.655 ms | 1 - 777 MB | NPU
122
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® SA8650P | 43.655 ms | 1 - 777 MB | NPU
123
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® SA8255P | 43.655 ms | 1 - 777 MB | NPU
124
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 64.25 ms | 1 - 962 MB | NPU
125
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® SA7255P | 136.992 ms | 1 - 765 MB | NPU
126
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® SA8295P | 50.824 ms | 1 - 629 MB | NPU
127
+ | NAFNet-DeNoise | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 20.348 ms | 1 - 764 MB | NPU
128
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® QCS9075 | 46.141 ms | 1 - 232 MB | NPU
129
+ | NAFNet-DeNoise | TFLITE | float | Qualcomm® QCS8750 | 20.348 ms | 1 - 764 MB | NPU
130
+
131
+ ## License
132
+ * The license for the original implementation of NAFNet-DeNoise can be found
133
+ [here](https://github.com/megvii-research/NAFNet/blob/main/LICENSE).
134
+
135
+ ## References
136
+ * [Simple Baselines for Image Restoration](https://arxiv.org/abs/2204.04676)
137
+ * [Source Model Implementation](https://github.com/megvii-research/NAFNet.git)
138
+
139
+ ## Community
140
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
141
+ * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
release_assets.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "0.55.0",
3
+ "precisions": {
4
+ "float": {
5
+ "universal_assets": {
6
+ "tflite": {
7
+ "tool_versions": {
8
+ "qairt": "2.45.0.260326154327",
9
+ "litert": "1.4.3"
10
+ },
11
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-tflite-float.zip"
12
+ },
13
+ "qnn_dlc": {
14
+ "tool_versions": {
15
+ "qairt": "2.45.0.260326154327"
16
+ },
17
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-qnn_dlc-float.zip"
18
+ },
19
+ "onnx": {
20
+ "tool_versions": {
21
+ "qairt": "2.42.0.251225135753_193295",
22
+ "onnx_runtime": "1.25.0"
23
+ },
24
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-onnx-float.zip"
25
+ }
26
+ }
27
+ },
28
+ "w8a16": {
29
+ "universal_assets": {
30
+ "qnn_dlc": {
31
+ "tool_versions": {
32
+ "qairt": "2.45.0.260326154327"
33
+ },
34
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-qnn_dlc-w8a16.zip"
35
+ },
36
+ "onnx": {
37
+ "tool_versions": {
38
+ "qairt": "2.42.0.251225135753_193295",
39
+ "onnx_runtime": "1.25.0"
40
+ },
41
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_denoise/releases/v0.55.0/nafnet_denoise-onnx-w8a16.zip"
42
+ }
43
+ }
44
+ }
45
+ }
46
+ }