-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathREADME.rtf
58 lines (56 loc) · 3.36 KB
/
README.rtf
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
{\rtf1\ansi\ansicpg1252\cocoartf1504\cocoasubrtf830
{\fonttbl\f0\fswiss\fcharset0 Helvetica;}
{\colortbl;\red255\green255\blue255;}
{\*\expandedcolortbl;;}
\paperw11900\paperh16840\margl1440\margr1440\vieww10800\viewh8400\viewkind0
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\f0\fs24 \cf0 The code in this package implements grayscale and color image denoising as described in the paper:\
\
Stamatis Lefkimmiatis\
Non-Local Color Image Denoising with Convolutional Neural Networks\
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, Hawaii, July 2017.\
\
Please cite the paper if you are using this code in your research.\
Please see the file LICENSE.txt for the license governing this code.\
\
\
Overview\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 \'97\'97\'97\'97\'97\'97\'97\'97\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 The function NLNET_DENOISE_DEMO demonstrates grayscale and color image denoising with the learned models from the paper, which can all be found in the folder \'93network/models\'94. The paper and supplementary material are provided in the folder "paper".\
\
The function NLNET_BSDS_VALIDATION_RESULTS can be used to obtain the results on the validation set extracted from the BSDS500 dataset for each one of the trained models. \
\
\b Note:
\b0 The results from this function differ slightly to the reported results in the CVPR paper since they are obtained with a slightly modified network architecture which provides on average a 0.1 dB PSNR increase in restoration quality for color images compared to the original network architecture. In particular the average results obtained with the original and the newest non-local networks are provided below \
\
Noise \uc0\u963 (std) | CNLNET_5x5 (original) | CNLNET_5x5 (new)\
15 33.69
\b 33.81
\b0 \
25 30.96
\b 31.08
\b0 \
50 27.64
\b 27.73
\b0 \
\
\
\
The folder \'93matlab/custom_layers\'94 contains all the CNN layers that are used to build the non-local networks described in the CVPR paper, while the folder \'93matlab/+misc\'94 includes some miscellaneous functions. The folder \'93matlab/custom_mex\'94 includes cpu and gpu mex files used to define some of the layers of the Non-local networks. \
\
Dependencies\
\'97\'97\'97\'97\'97\'97\'97\'97\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 The provided code has dependencies on the MatConvnet toolbox. The necessary functions are included in the folders \'93matlab/vl_layers\'94, \'93matlab/mex\'94, \'93matlab/src\'94 and \'93matlab/compatibility\'94. \
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 \
\
Contact\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 \'97\'97\'97\'97\'97\'97\'97\'97\
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0
\cf0 If you have questions, problems with the code, or found a bug, please let us know.\
Contact Stamatis Lefkimmiatis at [email protected]\
}