The left hand of UAB001 has a wedding ring:
${imageLink?synapseId=syn21479463&align=None&scale=100&responsive=true&altText=Just Married}
Training will have to work around this feature. It's not uncommon. I'm seeing rings also in UAB021, UAB022, and there are probably others.
Created by Lars Ericson lars.ericson Great. Thanks We remove the UAB296 set from the training data. Ahh... I see now.. you removed it..
I'm sorry I misread your reply..
My bad.. UAB296-LH.jpg which was talked about in this thread.
Please tell me which one you are looking for? Thanks. I posed that as a possible cause why I cannot find this image in the trainset.
Could you double check for me ?
Download the trainset and see if that image is present.
I tried many times and cannot find it.
We decided to make all training available to all. Please ignore those 50 sets of training. Hello dongmeisun,
I dont know but I :
1. redownloaded training.zip
2. extracted it
3. Sort by name.
4. This is what I get when I search for uab29*:
Was 296 perhaps part of the first 50 train samples ??
C:\Users\Gebruiker\Downloads\training\train>dir uab29*.*
Volume in drive C has no label.
Volume Serial Number is E489-D6CC
Directory of C:\Users\Gebruiker\Downloads\training\train
12/19/2019 11:34 AM 46,499 UAB297-LF.jpg
12/19/2019 11:34 AM 63,397 UAB297-LH.jpg
12/19/2019 11:34 AM 48,459 UAB297-RF.jpg
12/19/2019 11:34 AM 58,453 UAB297-RH.jpg
4 File(s) 216,808 bytes
0 Dir(s) 6,862,487,552 bytes free
@Juul de puul please make sure you got right training data since we removed UAB296 already. Thanks. @arielis
I cannot find UAB296-LH.jpg in the trainset..
Am I missing something here ? @arielis @allawayr I agree, and that's also what I just got from another organizer.
@lars.ericson We will keep the way it's. Please try to overcome.
Thanks all.
@dongmeisun - based on the number of films with jewelry it looks to be common practice to allow the patient to leave them on.
Therefore, it's probably important that algorithms should not get tripped up by the presence or absence of jewelry - I suggest we leave them in.
Personally I have succeeded to overcome the metallic artifacts in my segmentation algorithm, but there are images which quality is so degraded, that I was not able to use for training.
e.g. UAB296-LH.jpg
I hope there are no such instances in the test set...
@lars.ericson
Thanks for pointing these. I noticed the problem. Due to the availability of data I hesitated to remove them. If the rings will affect training, we have to move the sets of films. I will discuss with challenge organizers and let you all know our decision soon.