python src/deep/unet/unet_trainer.py config/unet_splits/split01.ini？
langbin2014 opened this issue · comments
nothing？why？Anyone can help me？
I found that there is only Class definition in src/deep/unet/unet_trainer.py . No running command. So nothing is done. There are something remained to be done src/deep/unet/unet_trainer.py . I did not have idea how to do it yet.
我也看到了，改了下，可以运行。O(∩_∩)O谢谢哈
@langbin2014 我也遇到了batch分母为零的错误！发现是这样的，default.ini里面的文件路径，需要在后面加上 _ALL 才匹配，加上之后就好了！
@JianqiaoAirport 谢谢，我试试
@langbin2014 发现后面还是有各种问题搞不定，如果你调好了能不能给@我我讲一下呀？谢谢！
have you guys figured out how to call src/deep/unet/unet_trainer.py config/unet_splits/split01.ini??
-
UNetTrainer.init() doesn't take a config path as a parameter so far.
-
The train() method in unet_trainer.py takes parameters 'train_splits, filenames_val, train_generator, val_generator'. How to properly set up these?
please advise.
Thanks!
looking at scripts/split01.sh, the training command should really be something like:
python train.py ../../config/titan_x_default.ini ../../config/split01.ini
although this seems to be in conflict with hard-coded line in params.py:
params = Params(['../../config/default.ini']+sys.argv[1:])
Really hope the author could clarify how to run a basic training procedure using the 10 subsets of data on CPU and GPU respectively...
Hi galactica147,
It's not in conflict with the line you found in params.py, it forms a list of the default configurations and the other script arguments, and loads them in that order (so your last configuration file overwrites earlier ones).
After the DSB competition is over (less than a month now), I can share the code of that. It features a 3D segmentation network for candidate generation, which I found to work quite a bit better. It's also faster in every way (A U-net implementation that is 10x as fast).
that'd be great @gzuidhof! Thanks and look forward to that
I met the same problem.
Have this issue been solved yet？
@langbin2014
@JianqiaoAirport
求问如何解决这个问题？
@JianqiaoAirport 如何修改路径,在路径中加_all阿? 谢谢阿!
How can I change the path in default.ini to avoid ZeroDivisionError? How to add "_all" to path? Thanks!
@gzuidhof how to run Unet training process?Thank you
@gzuidhof
1.In Preprocessing, I run src/data_processing/create_same_spacing_data_NODULE.py, but the error seems to tell me creat many dir first, such as '1_1_1mm_slices_lung/subset0' .However, I don't know how to use the 'set of segmentations of the lungs' that you talked about.
2.When I run ./src/deep/train.py config/unet_splits/split01.ini, error occur:
Loaded configurations from (in order) ['config/unet_splits/split01.ini']
Traceback (most recent call last):
File "./src/deep/train.py", line 4, in
from params import params as P
File "/home/ubuntu/Documents/project/luna16/luna16-master/src/deep/params.py", line 95, in
params = Params(['../../config/default.ini']+sys.argv[1:])
File "/home/ubuntu/Documents/project/luna16/luna16-master/src/deep/params.py", line 25, in init
self.PIXELS = cf.getint('dataset','pixels')
File "/home/ubuntu/anaconda2/lib/python2.7/ConfigParser.py", line 359, in getint
return self._get(section, int, option)
File "/home/ubuntu/anaconda2/lib/python2.7/ConfigParser.py", line 356, in _get
return conv(self.get(section, option))
File "/home/ubuntu/anaconda2/lib/python2.7/ConfigParser.py", line 618, in get
raise NoOptionError(option, section)
ConfigParser.NoOptionError: No option 'pixels' in section: 'dataset'
