search

jianpengz / DoDNet

[CVPR2021] DoDNet: Learning to segment multi-organ and tumors from multiple partially labeled datasets
GNU General Public License v3.0
177 stars 30 forks source link

KiTS19 #6

Closed ChiYe622 closed 3 years ago

ChiYe622 commented 3 years ago

Hi, you really did an excelent job, and thanks for your sharing. I have already prepared all datasets except KiTS19, other datasets work well on your code, they all got right results. I followed the steps in KiTS19, but this result run by evaluate.py and postp.py is wrong. So, would you please tell me more details about how to get right KiTS19? Thanks!

jianpengz commented 3 years ago

Did you re-space KiTS19 data by using the provided script re_spacing.py?

ChiYe622 commented 3 years ago

Yes, I re-space all datasets including KiTS19, but only KiTS19 got wrong result.

jianpengz commented 3 years ago

I didn't have this issue before. Did you try with our pre-trained model or train it by yourself? Could the model fit the training samples on KiTS? I guess this problem was caused by the data.

ChiYe622 commented 3 years ago

I used your pre-trained model, and I tried step by step two times, both failed. :-(

jianpengz commented 3 years ago

I tested KiTS using pre-trained model again and all cases work well during my testing. Here is a re-spaced case for test. Pls let me know whether it works in your environment. example

ChiYe622 commented 3 years ago

I run your re-space case_00000 and my re-space case_00000 in the same environment, these are the results. I run case_00000 only.

evaluate.py

post.py

jianpengz commented 3 years ago

Did you check is there a difference between two respacing cases?

kangmintong commented 3 years ago

Hi, you really did an excelent job, and thanks for your sharing. I have already prepared all datasets except KiTS19, other datasets work well on your code, they all got right results. I followed the steps in KiTS19, but this result run by evaluate.py and postp.py is wrong. So, would you please tell me more details about how to get right KiTS19? Thanks!

I got the same problem. Have you resolved the problem? Thanks.

jianpengz commented 3 years ago

Hi guys, The pre-trained model was trained on the spacing of 3x0.76x0.76 for kits. To reproduce the results on kits, pls re-spacing the data to 3x0.76x0.76 and then test. Another pre-trained model trained on the spacing of 1.5x0.8x0.8 for kits would be released later.

sherjy commented 3 years ago

Hiii @ChiYe622 , did you met this problem? #11 , this problem occurs when I run codes on LiTS dataset only. I checked the image shapes after 'MOTSDataSet' processing, finding some image shapes becomes wrong, e.g. preset resize to (batch, channel, 64, 192, 192), but it'll happen to be a wrong shape (batch,channel,64,146,192) or other wrong shapes sometimes. I wonder if this problem comes from the data reason or sth else? Thanks in advance!