search

cszn / DnCNN

Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising (TIP, 2017)
https://cszn.github.io/
1.42k stars 534 forks source link

data_size is NAN #18

Closed ghost closed 6 years ago

ghost commented 6 years ago

I met some difficuties after use vl_compilenn().

>> Demo_Train_model_64_25_Res_Bnorm_Adam
     layer|      0|      1|      2|      3|      4|      5|      6|      7|      8|      9|     10|     11|     12|     13|     14|     15|     16|     17|
      type|  input|   conv|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|
      name|    n/a| layer1| layer2| layer3| layer4| layer5| layer6| layer7| layer8| layer9|layer10|layer11|layer12|layer13|layer14|layer15|layer16|layer17|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   support|    n/a|      3|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|
  filt dim|    n/a|      1|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|
filt dilat|    n/a|      1|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|
 num filts|    n/a|     64|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|
    stride|    n/a|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
       pad|    n/a|      1|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   rf size|    n/a|      3|      3|      5|      5|      5|      7|      7|      7|      9|      9|      9|     11|     11|     11|     13|     13|     13|
 rf offset|    n/a|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
 rf stride|    n/a|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
 data size|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|
data depth|    NaN|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|
  data num|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
  data mem|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|
 param mem|    n/a|    2KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|

     layer|     18|     19|     20|     21|     22|     23|     24|     25|     26|     27|     28|     29|     30|     31|     32|     33|     34|     35|
      type|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|
      name|layer18|layer19|layer20|layer21|layer22|layer23|layer24|layer25|layer26|layer27|layer28|layer29|layer30|layer31|layer32|layer33|layer34|layer35|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   support|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|
  filt dim|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|
filt dilat|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|
 num filts|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|
    stride|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
       pad|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   rf size|     15|     15|     15|     17|     17|     17|     19|     19|     19|     21|     21|     21|     23|     23|     23|     25|     25|     25|
 rf offset|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
 rf stride|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
 data size|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|
data depth|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|
  data num|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
  data mem|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|
 param mem|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|

     layer|     36|     37|     38|     39|     40|     41|     42|     43|     44|     45|     46|     47|     48|     49|
      type|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|  bnorm|   relu|   conv|   loss|
      name|layer36|layer37|layer38|layer39|layer40|layer41|layer42|layer43|layer44|layer45|layer46|layer47|layer48|layer49|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   support|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|      1|      3|      1|
  filt dim|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|
filt dilat|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|    n/a|      1|    n/a|
 num filts|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|     64|    n/a|    n/a|      1|    n/a|
    stride|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
       pad|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|      0|      1|      0|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
   rf size|     27|     27|     27|     29|     29|     29|     31|     31|     31|     33|     33|     33|     35|     35|
 rf offset|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
 rf stride|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|      1|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
 data size|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|NaNxNaN|
data depth|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|     64|      1|      1|
  data num|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|    128|      1|
----------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
  data mem|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|    NaN|
 param mem|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|  144KB|    1KB|     0B|    2KB|     0B|
ghost commented 6 years ago

First step, I run GenerateTrainingPatches.m,Second step ,I run Demo_Train_model_64_25_Res_Bnorm_Adam.m to train.But I find data_size is NAN.what's wrong with it? then I should run Demo_Test_model_64_25_Res_Bnorm_Adam.m?

ghost commented 6 years ago

By the way,if I just want to validate your code,could I run Demo_test_DnCNN.m directly? But PSNR and SSIM both are NAN.so,could you figure the problem?

ghost commented 6 years ago

maybe I solved above problem,but I met the follow problem :

>> Demo_test_DnCNN
引用了不存在的字段 'dilate'。

出错 vl_simplenn (line 303)
        'dilate', l.dilate, ...

出错 Demo_test_DnCNN (line 64)
    res    = vl_simplenn(net,input,[],[],'conserveMemory',true,'mode','test');

maybe I solved above problem,but

cszn commented 6 years ago

Train400

See https://github.com/cszn/DnCNN/issues/11

chenjian123321 commented 6 years ago

hello , I run Demo_test_DnCNN.m , But PSNR and SSIM both are NAN.so,could you figure the problem? and run DnCNN_train ,the result have error and is Input factor is insufficient DnCNN_train(line 75) net = vl_simplenn_tidy(net); can you help me

ngcthuong commented 6 years ago

I dont have this problem. It seems like you might have some problem with your training. How bout your training error? it should reduce around 1.0.

chenjian123321 commented 6 years ago

what is difference for 1.0 and 1.1? l run the 1.0 also have error.

chenjian123321 commented 6 years ago

l first run Demo_test_DnCNN3.m, the result is ----------------------------------------------- ----BSD68------Gaussian Denoising-----

Average PSNR is NaNdB Average SSIM is NaN

----Set5-----Super-Resolution-----

Average PSNR is NaNdB Average SSIM is NaN

----Set14-----Super-Resolution-----

Average PSNR is NaNdB Average SSIM is NaN

----BSD100-----Super-Resolution-----

Average PSNR is NaNdB Average SSIM is NaN

----Urben100-----Super-Resolution-----

Average PSNR is NaNdB Average SSIM is NaN

----classic5------Deblocking-----

Average PSNR is NaNdB Average SSIM is NaN

----LIVE1------Deblocking-----

Average PSNR is NaNdB Average SSIM is NaN

ngcthuong commented 6 years ago

For the NAN problem in version 1.1, you should (1) add the path to 'train400' data, then (2) run 'data\GenerateTrainingPatches.m', so that it will generate the training data for you. The training data is storaged in 'data\TrainingPatches\imdb_40_128.mat'. And please check your training data size 'imdb_40_128.mat' whether you really have the data or not.

chenjian123321 commented 6 years ago

thank you very much

chenjian123321 commented 6 years ago

wait , what is (1), 1.1 add or 1.0 add?

ngcthuong commented 6 years ago

You should add it for TrainingCodes_v1.1 since it doesn't have 'Train400' images in their 'data' folder. The 'Train400' images path is located in 'TrainingCodes_v1.0' folder.

chenjian123321 commented 6 years ago

OK, thank you

moinul95 commented 6 years ago

Hi, I don't understand how the mat files in model are created. Can you please tell in what sequence the code should run?

chenjian123321 commented 6 years ago

hello. first, you should run Demo_test_DnCNN3. if you use GPU, ni should remove "%net = vl_simplenn_tidy(net);" "%", and if have error, in Demo_test_DnCNN3 add your vl_setupnn path. then run GenerateData_model_64_25_Res_Bnorm_Adam.

moinul95 commented 6 years ago

I am actually new to deep learning. I didn't get how the models are created as .mat files. We need to run some code and then save it as mat file. Where is this code to create model? Thank you.

cszn commented 6 years ago

@moinul95 https://github.com/cszn/DnCNN/blob/52faf8897d7b9f9e22803df39a3606fbc8c04ac1/TrainingCodes/DnCNN_TrainingCodes_v1.1/DnCNN_train.m#L115

allenkate12 commented 6 years ago

I am new to deep learning. how did u you resolve the data_size and data_meme NaN issue? @goonder .

YananXM commented 5 years ago

也许我解决了上面的问题，但我遇到了以下问题：

>> Demo_test_DnCNN
引用了不存在的字段 'dilate'。

出错 vl_simplenn (line 303)
        'dilate', l.dilate, ...

出错 Demo_test_DnCNN (line 64)
    res    = vl_simplenn(net,input,[],[],'conserveMemory',true,'mode','test');

也许我解决了上面的问题，但是

您好，我现在也遇到了同样的问题，请问您是怎么解决的。

15230127713 commented 5 years ago

I am new to deep learning. how did you resolve the data_size and data_meme NaN issue? @goonder @allenkate12

chenjian123321 commented 5 years ago

First, please check your CUDN and cudnn is downloaded. Second, check your GPU memory. -----Original Message----- From: "15230127713"notifications@github.com To: "cszn/DnCNN"DnCNN@noreply.github.com; Cc: "chenjian123321"chenjian105@naver.com; "Comment"comment@noreply.github.com; Sent: 2019-05-10 (금) 10:09:39 (GMT+09:00) Subject: Re: [cszn/DnCNN] data_size is NAN (#18)

I am new to deep learning. how did you resolve the data_size and data_meme NaN issue? @goonder @allenkate12 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

1804941502 commented 1 year ago

maybe I solved above problem,but I met the follow problem :

>> Demo_test_DnCNN
引用了不存在的字段 'dilate'。

出错 vl_simplenn (line 303)
        'dilate', l.dilate, ...

出错 Demo_test_DnCNN (line 64)
    res    = vl_simplenn(net,input,[],[],'conserveMemory',true,'mode','test');

maybe I solved above problem,but

How can you solve this problem,could you tell me? I have met the same problem,my datasiaze is also NAN.

YananXM commented 1 year ago

你好，你的信件我已收到，谢谢，薛亚楠。