Tseng Kuan Lun

You can using predicted segmentation with different color (each label with distinct color).

For example: ls dataset/BRATS2018/training/HGG/Brats18_2013_10_1/ => Brats18_2013_10_1_flair.nii.gz Brats18_2013_10_1_seg.nii.gz Brats18_2013_10_1_t1ce.nii.gz Brats18_2013_10_1_t1.nii.gz Brats18_2013_10_1_t2.nii.gz You can also modify dataloader.py to suit your need.

Do you use windows ? If yes you should implement your data loader since zmq doesn’t support windows. You can check tensorpack for more detail kongbo96 於 2019年8月17日 週六，下午3:27寫道： >...

I used the official implementation from tensorflow models. rupalkapdi 於 2019年1月22日 週二，下午5:00寫道： > Hello, I have gone through your paper 'End-to-End Cascade Network for 3D > Brain Tumor Segmentation in...

I didn'y met this problem before. 60G memory usage sounds impossible to me. Can you try to use offline evaluation to see if the problem still exist ? ex: python3...

The warning is as expected, variable global_step:0, learning_rate:0 are only used in training mode. Is there other exception ?

Nice ! It will be nice if you submit a pull request ! Maybe other people are facing the problem.

@mini-Shark Your problem is that you cannot do evaluation and training at the same time because of memory bottleneck ? You can try to change config. NO_CACHE = True to...

@huangmozhilv Thanks ! I will try to investigate tensorpack source code to figure out a workaround. The ugly solution is that you discard the queue input and just use feed_dict.

The error shows that input has error. You can debug the input pipeline maybe dir not exist ? Galvin 於 2019年1月21日 週一，下午4:41寫道： > thank you open source! when i run...