mkfmiku / raisr

Google Rapid and Accurate Image Super Resolution

RAISR

A unofficial Python implementation of Google Rapid and Accurate Image Super Resolution

Prepare

Dataset will use 291

./getdataset.sh
python train.py

Train

usage: train.py [-h] [--rate RATE] [--patch PATCH] [--Qangle QANGLE]
                [--Qstrength QSTRENGTH] [--Qcoherence QCOHERENCE]
                [--datasets DATASETS]

RAISR

optional arguments:
  -h, --help            show this help message and exit
  --rate RATE           upscale scale rate
  --patch PATCH         image patch size
  --Qangle QANGLE       Training Qangle size
  --Qstrength QSTRENGTH
                        Training Qstrength size
  --Qcoherence QCOHERENCE
                        Training Qcoherence size
  --datasets DATASETS   path save the train dataset

Todo:

• left test.py implment to get PSNR/SSIM/Runtime

raisr's Issues

encountered an OverflowError when running train.py

I've add the line in the other issue and then i run the train.py. There i get an OverflowError: cannot convert float infinity to integer in the line:
File "/home/hatai/code/tensorflow/picZip/RAISR-master/model/hashTable.py", line 29, in hashTable
return int(angle),int(strength),int(coherence)
Anyone knows how to fix it? Thanks a lot

Error

Traceback (most recent call last):
File "train.py", line 44, in
LR = cv2.GaussianBlur(LR,(0,0),2);
NameError: name 'LR' is not defined

New Error

y, cb, cr = lr.split()
AttributeError: 'Array' object has no attribute 'split'

and integer or boolean arrays are valid indices

Hi friend, thanks for the project,

I know that your time is valuable, but I would appreciate it too much if you could guide me,

I am trying to run the file train.py in python 3.5, it shows me the following error:
File "train.py", line 57, in
Q[angle,strength,coherence] = Q[angle,strength,coherence] + patch.T*patch
IndexError: only integers, slices (:), ellipsis (...), numpy.newaxis (None) and integer or boolean arrays are valid indices

Beforehand thank you very much,

Greetings from Colombia

Excuse my English I use the google translator

some doubts about your code

hi,dear author
recently,i have read this paper,and luckily i have found your code.but i have some question about it.
1.i guess you wrote your code in python 3.0,because i can't run it in python2.7，then i changed some codes,and succeed.
2.i have some doubts about your codes in reconstruction your low solution image.
the following are some codes of you:
[angle,strength,coherence] = hashTable(patch,Qangle,Qstrenth,Qcoherence) f = H[angle,strength,coherence] patch = patch.reshape(1,-1) LRImage[jPixel,iPixel] = patch*np.matrix(f).T jPixel = jPixel+1

in google's paper,after we classified our patches into buckets,there may be no patch or a little patches in buckets because there are often many more horizontal and vertical structures in imagery,and flat regions are common,but i can't see what solution you make in your paper.

为了表述更清楚，我下面使用中文（见笑）：在谷歌论文分桶的时候，有些结构经常出现有些出现得少使得桶子里面的图像块数非常少，你是怎样处理这个的？

my email is ：[email protected]
i am looking forward your reply!

create folders to store LR / HR seperately

seems like the codes downscale and upscale in train.py at real time. Suggest to do it offline. FFMPEG has different scaling filter. we can reuse that to generate training picture. I can give a hand if you want.

Problem with test.py

Hi, can you please have a look at your test.py -- I don't think it is working properly. Looks like you are reading images separately from a directory and then again you are relying on image read from the training set. I understand you are trying to show the comparison but LR is not defined if we comment out after line 37.

Error when run ./train.py

when I run with python3 ./train.py, I got an error:

HashMap of ./datasets/BSDS200/100075.png

Traceback (most recent call last):
File "./train.py", line 81, in
main()
File "./train.py", line 68, in main
Q[j, t] += A * A.T
ValueError: operands could not be broadcast together with shapes (121,121) (66,66) (121,121)

just

Error :S

I keep getting this error:

StephenPereira@tuwigmlt7nbymh2 ~/RAISR$ python train.py
Namespace(Qangle=24, Qcoherence=3, Qstrength=3, datasets='./datasets/291/', patch=11, rate=3)
('Load dataset ', 161)
  0%|                                                                                                                                                                                                                                                   | 0/161 [00:00<?, ?it/s]HashMap of ./datasets/291/ACC_ 974959_THA_33.jpg

Traceback (most recent call last):
  File "train.py", line 80, in <module>
    main()
  File "train.py", line 40, in main
    im = misc.imread(image_path, mode='YCbCr')
AttributeError: 'module' object has no attribute 'imread'