Open StefanDenn3r opened 8 months ago
Furthermore, there are quite some samples which are just empty:
"ee4d421f59bd462d212ce24753493da4": [
"US/thyroid/usn418235.png",
"CT/abd/normal/abd-normal039513.png",
"CT/abd/normal/abd-normal063889.png",
"CT/abd/normal/abd-normal053270.png",
"CT/abd/normal/abd-normal002528.png",
"CT/abd/normal/abd-normal046244.png",
"CT/abd/normal/abd-normal016491.png",
"CT/abd/normal/abd-normal056338.png",
"CT/abd/normal/abd-normal016490.png",
"CT/abd/normal/abd-normal068850.png",
"CT/abd/normal/abd-normal047066.png",
"CT/abd/normal/abd-normal049461.png",
"CT/abd/normal/abd-normal039070.png",
"CT/abd/normal/abd-normal018254.png",
"CT/abd/normal/abd-normal017359.png",
"CT/abd/normal/abd-normal063883.png",
"CT/abd/normal/abd-normal017361.png",
"CT/abd/normal/abd-normal016112.png",
"CT/abd/normal/abd-normal021904.png",
"CT/abd/normal/abd-normal051763.png",
"CT/abd/normal/abd-normal014918.png",
"CT/abd/normal/abd-normal035602.png",
"CT/abd/normal/abd-normal034951.png",
"CT/abd/normal/abd-normal016113.png",
"CT/abd/normal/abd-normal068847.png",
"CT/abd/normal/abd-normal069026.png",
"CT/abd/normal/abd-normal020266.png",
"CT/abd/normal/abd-normal058409.png",
"CT/abd/normal/abd-normal016492.png",
"CT/abd/normal/abd-normal014914.png",
"CT/abd/normal/abd-normal016117.png",
"CT/abd/normal/abd-normal068845.png",
"CT/abd/normal/abd-normal024872.png",
"CT/abd/normal/abd-normal063887.png",
"CT/abd/normal/abd-normal006929.png",
"CT/abd/normal/abd-normal038923.png",
"CT/abd/normal/abd-normal016489.png",
"CT/abd/normal/abd-normal014920.png",
"CT/abd/normal/abd-normal068849.png",
"CT/abd/normal/abd-normal002527.png",
"CT/abd/normal/abd-normal023009.png",
"CT/abd/normal/abd-normal028309.png",
"CT/abd/normal/abd-normal068848.png",
"CT/abd/normal/abd-normal012402.png",
"CT/abd/normal/abd-normal002529.png",
"CT/abd/normal/abd-normal027692.png",
"CT/abd/normal/abd-normal014917.png",
"CT/abd/normal/abd-normal038921.png",
"CT/abd/normal/abd-normal063886.png",
"CT/abd/normal/abd-normal017360.png",
"CT/abd/normal/abd-normal014922.png",
"CT/abd/normal/abd-normal016493.png",
"CT/abd/normal/abd-normal063884.png",
"CT/abd/normal/abd-normal048245.png",
"CT/abd/normal/abd-normal026267.png",
"CT/abd/normal/abd-normal050842.png",
"CT/abd/normal/abd-normal068196.png",
"CT/abd/normal/abd-normal059727.png",
"CT/abd/normal/abd-normal004505.png",
"CT/abd/normal/abd-normal039068.png",
"CT/abd/normal/abd-normal025695.png",
"CT/abd/normal/abd-normal043546.png",
"CT/abd/normal/abd-normal066519.png",
"CT/abd/normal/abd-normal038082.png",
"CT/abd/normal/abd-normal045106.png",
"CT/abd/normal/abd-normal030263.png",
"CT/abd/normal/abd-normal066054.png",
"CT/abd/normal/abd-normal002526.png",
"CT/abd/normal/abd-normal054393.png",
"CT/abd/normal/abd-normal034294.png",
"CT/abd/normal/abd-normal039072.png",
"CT/abd/normal/abd-normal002524.png",
"CT/abd/normal/abd-normal019419.png",
"CT/abd/normal/abd-normal007039.png",
"CT/abd/normal/abd-normal020776.png",
"CT/abd/normal/abd-normal018253.png",
"CT/abd/normal/abd-normal014921.png",
"CT/abd/normal/abd-normal011770.png",
"CT/abd/normal/abd-normal050088.png",
"CT/abd/normal/abd-normal039073.png",
"CT/abd/normal/abd-normal058178.png",
"CT/abd/normal/abd-normal056852.png",
"CT/abd/normal/abd-normal063888.png",
"CT/abd/normal/abd-normal048689.png",
"CT/abd/normal/abd-normal017358.png",
"CT/abd/normal/abd-normal016115.png",
"CT/abd/normal/abd-normal002651.png",
"CT/abd/normal/abd-normal014915.png",
"CT/abd/normal/abd-normal040704.png",
"CT/abd/normal/abd-normal058408.png",
"CT/abd/normal/abd-normal002530.png",
"CT/abd/normal/abd-normal038924.png",
"CT/abd/normal/abd-normal038770.png",
"CT/abd/normal/abd-normal063005.png",
"CT/abd/normal/abd-normal012403.png",
"CT/abd/normal/abd-normal005817.png",
"CT/abd/normal/abd-normal038922.png",
"CT/abd/normal/abd-normal039071.png",
"CT/abd/normal/abd-normal059091.png",
"CT/abd/normal/abd-normal048244.png",
"CT/abd/normal/abd-normal016116.png",
"CT/abd/normal/abd-normal069025.png",
"CT/abd/normal/abd-normal039069.png",
"CT/abd/normal/abd-normal025111.png",
"CT/abd/normal/abd-normal060343.png",
"CT/abd/normal/abd-normal021311.png",
"CT/abd/normal/abd-normal018255.png",
"CT/abd/normal/abd-normal067181.png",
"CT/abd/normal/abd-normal014919.png",
"CT/abd/normal/abd-normal064379.png",
"CT/abd/normal/abd-normal011865.png",
"CT/abd/normal/abd-normal048784.png",
"CT/abd/normal/abd-normal029406.png",
"CT/abd/normal/abd-normal063885.png",
"CT/abd/normal/abd-normal016114.png",
"CT/abd/normal/abd-normal062317.png",
"CT/abd/normal/abd-normal068846.png",
"CT/abd/normal/abd-normal054395.png",
"CT/abd/normal/abd-normal054303.png",
"CT/abd/normal/abd-normal002400.png",
"CT/abd/normal/abd-normal044746.png",
"CT/abd/normal/abd-normal068195.png",
"CT/abd/normal/abd-normal066049.png",
"CT/abd/normal/abd-normal033146.png",
"CT/abd/normal/abd-normal068705.png",
"CT/abd/normal/abd-normal014916.png",
"CT/abd/normal/abd-normal057421.png",
"CT/abd/normal/abd-normal051215.png",
"CT/abd/normal/abd-normal065509.png",
"CT/abd/normal/abd-normal009734.png",
"CT/abd/normal/abd-normal032744.png",
"CT/abd/normal/abd-normal022428.png",
"CT/abd/normal/abd-normal028845.png",
"CT/abd/normal/abd-normal028748.png",
"CT/abd/normal/abd-normal063890.png",
"MR/mriabd/normal/mri-abd-normal047494.png",
"MR/mriabd/normal/mri-abd-normal047532.png",
"MR/mriabd/normal/mri-abd-normal047530.png",
"MR/mriabd/normal/mri-abd-normal047531.png",
"MR/mriabd/normal/mri-abd-normal047529.png",
"MR/mriabd/normal/mri-abd-normal047492.png",
"MR/mriabd/normal/mri-abd-normal047493.png",
"MR/mriabd/normal/mri-abd-normal047491.png",
"MR/af/chondral_abnormality/ankle026697.png"
],
Hi, first of all, thanks for publishing the RadImageNet dataset! While working with it, I discovered that there are quite some duplicate entries, when checking the MD5 hash of the files.
CT/lung/interstitial_lung_disease/lung009382.png
andCT/lung/Nodule/lung009382.png
(Note: same filename)MR/af/Plantar_plate_tear/foot040499.png
andMR/af/plantar_fascia_pathology/ankle027288.png
(Note: different filename)MR/af/hematoma/foot079779.png
andMR/af/hematoma/ankle053088.png
US/gb/usn309850.png
andUS/gb/usn309851.png
US/ovary/usn326815.png
andUS/kidney/usn348701.png
So far, I haven't checked if the duplicates are across your utilized dataset split, but since you write in your paper that you split patient wise, this shouldn't be the case.
However, the following questions arise:
Since, from my understanding of the paper, this dataset is intended as a single-label, not a multi-label dataset, I am confused to find samples as in the first case. Now the question arises, can the dataset be considered as a multi-label dataset where all 165 pathologies are labeled in all images if present?
For the cases 2.-4. those duplicates are just creating an imbalance but don't provide additional information. Are you planning to remove them?
In total this results in: Number of duplicate groups: 62751 Total duplicate files: 126074
I attached a duplicates.json with all the duplicates found. It's a dictionary where each key is a MD5 hash and its value is a list of image paths with that hash.
Here is the script I wrote to detect the duplicates, to ensure reproducibility.