Multicue dataset

[WCJ1998]

Hello, could you provide a link to the Multicue dataset and the enhanced data? thank you very much

[rrryan2016]

Hello, could you provide a link to the Multicue dataset and the enhanced data? thank you very much

1. Link of Multicue: https://serre-lab.clps.brown.edu/resource/multicue/
2. You can take a reference to https://github.com/yun-liu/RCF/issues/124 for the data augmentation.

[WCJ1998]

Hello, the downloaded Multicue dataset is all pictures. How do you use it when testing the ODS target value? Because the labels on the BSDS dataset are in mat format, can you tell me how the multicue dataset is tested? thank you very much!

[rrryan2016]

Transfer the png files to mat files.

Take the MATLAB codes below for reference:

main_path =  'C:\Personal\data\multicue_groundtruth\combine\edge_input'; 
output_main_path =  'C:\Personal\data\multicue_groundtruth\combine\edge_output'; 
image_names = {'4580_left_0165',  '4680_left_0170', '4176_left_0273',  '4486_left_0036', '4163_left_0052',  '4491_left_0070', '4582_left_0020',  '4673_left_0115', '4108_left_0948',  '4588_left_0200', '4179_left_0315',  '4676_left_0093', '4569_left_0060',  '4175_left_0184', '4592_left_0020',  '4571_left_0062', '4108_left_2343',  '4457_left_0050', '4187_left_0071',  '4113_left_2396', '4171_left_0221',  '4492_left_0090', '4383_left_0010',  '4596_left_0020', '4378_left_0011',  '4406_left_0030', '4568_left_0020',  '4178_left_0052', '4186_left_0127',  '4494_left_0070', '4386_left_0020',  '4586_left_0080', '4523_left_0070',  '4396_left_0050', '4108_left_0082',  '4689_left_0050', '4184_left_0048',  '4209_left_0034', '4453_left_0020',  '4640_left_0080', '4665_left_0040',  '4524_left_0040', '4558_left_0260',  '4097_left_0667', '4668_left_0130',  '4188_left_0315', '4177_left_0071',  '4200_left_0221', '4687_left_0130',  '4548_left_0080', '4108_left_0446',  '4470_left_0010', '4404_left_0010',  '4663_left_0010', '4625_left_0140',  '4504_left_0025', '4449_left_0290',  '4475_left_0010', '4113_left_0082',  '4407_left_0100', '4567_left_0030',  '4420_left_0040', '4603_left_0050',  '4385_left_0020', '4398_left_0100',  '4380_left_0025', '4669_left_0080',  '4402_left_0040', '4403_left_0010',  '4451_left_0020', '4597_left_0010',  '4210_left_0296', '4172_left_0067',  '4114_left_0611', '4674_left_0100',  '4484_left_0040', '4483_left_0170',  '4552_left_0080', '4561_left_0100',  '4605_left_0220', '4190_left_0161',  '4101_left_0382',  '4609_left_0080', '4684_left_0010',  '4710_left_0035', '4220_left_0221',  '4409_left_0020', '4629_left_0050',  '4382_left_0015', '4638_left_0070',  '4685_left_0045', '4208_left_0052',  '4452_left_0010', '4192_left_0221',  '4377_left_0016', '4564_left_0010',  '4198_left_0034', '4100_left_2268',  '4644_left_0120', '4641_left_0030'};
for image=image_names
sprintf('processing %s ...',char(image))
for i=1:6 
    b =  imread(fullfile(main_path,sprintf('%s.%d.png',char(image),i)));
    b = logical(b);
    groundTruth{i}.Boundaries = b;
     save(fullfile(output_main_path,sprintf('%s.mat',char(image))),'groundTruth');
end
end

[WCJ1998]

I'm sorry to bother you again. I was using matlab to test the results of the multicue data set and found that it took more than ten hours to test an epoch on my computer, but it was not completed. I would like to ask if you have encountered this kind of problem. I still need to do other operations during the test. Thank you very much.

[rrryan2016]

Same problem in experiments. Simply because the resolution of Multicue (1280 by 720 pixels) is much larger than NYUD or BSDS500.

I found an evaluation code by Python (https://github.com/Walstruzz/edge_eval_python) but has not been verified yet. You can have a try if it can be more cost-efficient.

[WCJ1998]

thank you very much. This code is available, and it is significantly faster on my server. The result is the same as using matlab, but the code should only run on the cpu of the linux system, and the speed may depend on the cpu performance of the server.

[rrryan2016]

Bravo! Thanks for this kind sharing.

[WCJ1998]

Sorry to interrupt you again, I also tried data augmentation on the multicue dataset and trained it in the same way as BSDS500, but the final result was very poor. Would it be convenient for you to send me your data set and the final contour map? thank you very much

[rrryan2016]

guess @zhuoinoulu can give a hand :)

[zhuoinoulu]

@WCJ1998 @rrryan2016 the dataset can be downloaded at https://drive.google.com/file/d/1-tyt_KyzlYc9APafdh5mHJzh2K_F2hM8/view?usp=sharing

[XRONHH]

I'm sorry to bother you, i can't download your multicue dataset link https://github.com/zhuoinoulu/pidinet/issues/11#issuecomment-987661561 ,due to network reasons. Could you give me a domestic link?this is my e-mail:529948572@qq.com