Images have induced borders in the JPG. These can be due to centering or
For example:
UAB039-LH
UAB039-RH
UAB046-LF
UAB046-LH
UAB046-RF
UAB088-RF
Probably one of the best cases is UAB109-RH.
Also, artifacts such as that in UAB059-LH, what happens there?
Also - having UAB111-LH
Created by John Muschelli muschellij2
@muschellij2
What you pointed is right. For UAB195-LH, the cut part is not scored. I cut it to remove the bright edge.
All training data including 373 sets of films should be available soon. I had to remove some sets of films due to many defects. Please let us know if you have any suggestions. Please feel free to send via the Synapse email relay to:
LouBridges@synapse.org
dongmeisun@synapse.org
(Just a note, you have to send from your Synapse-registered email account, otherwise, the relay will bounce it back).
Thanks!
Robert Similarly, some of the fixes actually cause issues. For example version 1 has the full hand: UAB195-LH, but version 2 cuts off the middle finger top joint. How can these be uploaded to show the differences to address? I would upload to show you what I mean but don't want to redistribute these scans. Hi there @muschellij2 ,
I believe that the challenge steering committee decided to release the full training set based on feedback from you and others.
@dongmeisun/@LouBridges are working to finalize this set and we'll be releasing that when they are done, so that should eliminate this issue.
cheers,
Robert
So it seems that these are missing from the second data release but were in the first data release:
[1] "UAB245-LF.jpg" "UAB245-LH.jpg" "UAB245-RF.jpg" "UAB245-RH.jpg"
[5] "UAB344-LF.jpg" "UAB344-LH.jpg" "UAB344-RF.jpg" "UAB344-RH.jpg"
Also, these are the cases that have been added to the second version that were not in the first.
[1] "UAB061-LF.jpg" "UAB061-LH.jpg" "UAB061-RF.jpg" "UAB061-RH.jpg"
[5] "UAB166-LF.jpg" "UAB166-LH.jpg" "UAB166-RF.jpg" "UAB166-RH.jpg"
Thus, some people who have access in the 1st and second release have slightly more information than those who only have access to the second set. @muschellij2 thank you for pointing these out. @dongmeisun is working on correcting these. @LouBridges @dongmeisun Here are some other examples of data that could (perhaps) be cleaned.
Thanks for summarizing your findings, @muschellij2 ! UAB333-LH and UAB333-RF - arrows drawn onto radiograph.
Overall, here are the issues that I have seen with the data.
UAB196-LF - zoomed out
UAB245-LF - poor quality and not reoriented
UAB245-LH - poor quality and not reoriented
UAB245-RF - poor quality and not reoriented, and flipped top/bottom
UAB245-RH - poor quality and not reoriented
UAB333-LF - flipped top/bottom
UAB333-LH - flipped top/bottom
UAB333-RF - flipped top/bottom
UAB333-RH - flipped top/bottom
UAB344-LF - flipped top/bottom
UAB344-LH - flipped top/bottom
UAB344-RF - flipped top/bottom
UAB344-RH - flipped top/bottom, also markers
UAB352 and UAB357 all images fliipped
UAB363-LF - flipped
UAB422-LF - zoom
UAB422-RF - zoom and flip
UAB475 - flipped
UAB481-LF - flipped
UAB545-LH - both hands present
UAB545-RH - both hands present
UAB597-RH - "Sizes are approximate"
UAB628-RF - flipped