thanosvlo/MARL-for-Anatomical-Landmark-Detection

Data loader questions

Closed this issue · 9 comments

xychenunc commented

Hi,

When I run your code using the provided data, I cannot run the code without any errors. And I also want to know how many landmarks are detected at the same time? It seems you skipped the first 4 landmarks, why? (see the following code in dataReader.py)

def getLandmarksFromVTKFile(file):
''' 0-2 RV insert points
1 -> RV lateral wall turning point
3 -> LV lateral wall mid-point
4 -> apex
5-> center of the mitral valve
'''
with open(file) as fp:
landmarks = []
for i, line in enumerate(fp):
print('i: ', i, 'line: ', line)
if i == 5:
landmarks.append([float(k) for k in line.rstrip().split(',')])
elif i == 6:
landmarks.append([float(k) for k in line.rstrip().split(',')])
elif i > 6:
landmarks = np.asarray(landmarks).reshape((-1,3))
landmarks[:,[0, 1]] = -landmarks[:,[0, 1]]
return landmarks

thanosvlo commented

First of all , I would appreciate it if you were more polite... - changed the title of the issue

Regarding the landmarks the first 4 were skipped because the study focused on some specific ones, not all.
The user determines the amount of landmarks to be detected.

xychenunc commented

I feel so sorry about that.

I want to know how many landmarks can be detected at the same time? That is, given the shared part of the model, how many agents can be used (on a 12 GB gpu)? And I also want to know the speed and performance of this method, because I am not able to run the code successfully right now.

Thank you!

thanosvlo commented

I have successfully trained and inferred using up to 5 agents using a 16GB gpu. Inference is at 20-25fps depending on the hardware ive tested it with
As i said before you chose how many landmarks you want to detect , I've tried up to 5, no reason why not have more

xychenunc commented
thanosvlo commented

Training usually was done in 3-4 hours , inference is 20-25fps for a single MRI volume

xychenunc commented
thanosvlo commented

thats a tensorpack issue, You should check the version of tensorpack you have, also use python 3 not 2.7

xychenunc commented
yeloiss commented

I followed https://github.com/amiralansary/tensorpack-medical to install the same version of the dependencies. However, same problem still exists. Can you help look into this issue? Thanks so much!
…
________________________________ From: thanosvlo notifications@github.com Sent: Tuesday, August 13, 2019 2:22 PM To: thanosvlo/MARL-for-Anatomical-Landmark-Detection MARL-for-Anatomical-Landmark-Detection@noreply.github.com Cc: Chen, Xiaoyang xychen@email.unc.edu; Author author@noreply.github.com Subject: Re: [thanosvlo/MARL-for-Anatomical-Landmark-Detection] Data loader questions (#1) thats a tensorpack issue, You should check the version of tensorpack you have, also use python 3 not 2.7 — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#1?email_source=notifications&email_token=AHVAPZZU4BYBQDXNXICMNATQEL3WVA5CNFSM4ILMMNI2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4GREMQ#issuecomment-520950322>, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AHVAPZ4NYPFKM34UKMKNJTTQEL3WVANCNFSM4ILMMNIQ.

hello!
I also got the following error, so have you solved it?
"
raise NotImplementedError()
NotImplementedError
"
thx a lot!