FileNotFoundError: [Errno 2] No such file or directory: 'list_of_test_filenames.txt'

[crypdick]

Hi, I think this is a really cool project and I'm trying to use it in my research. I am trying to reproduce your results but I run into trouble because I don't have the training data:

 ~/bin/tensorpack-medical/examples/LandmarkDetection/DQN   master ●  python DQN.py --algo DQN --gpu 0
WARN: gym.spaces.Box autodetected dtype as <class 'numpy.uint8'>. Please provide explicit dtype.
Traceback (most recent call last):
  File "DQN.py", line 205, in <module>
    screen_dims=IMAGE_SIZE)
  File "/home/ubuntu/bin/tensorpack-medical/examples/LandmarkDetection/DQN/medical.py", line 157, in __init__
    self.files = filesListFetalUSLandmark(directory,files_list)
  File "/home/ubuntu/bin/tensorpack-medical/examples/LandmarkDetection/DQN/sampleTrain.py", line 321, in __init__
    self.files_list = [line.split('\n')[0] for line in open(files_list)]
FileNotFoundError: [Errno 2] No such file or directory: 'list_of_test_filenames.txt'

Where can I get the dataset? I see in your paper that you cite another paper's dataset, but I did not see a link to that dataset anywhere in that paper, either.

[crypdick]

Same issue with AutomaticViewPlanning:

  File "/home/ubuntu/bin/tensorpack-medical/examples/AutomaticViewPlanning/DQN/sampleTrain.py", line 244, in __init__
    self.files_list = [line.split('\n')[0] for line in open(files_list)]
FileNotFoundError: [Errno 2] No such file or directory: 'list_of_test_filenames.txt'

[amiralansary]

sampleTrain.py was written specifically for the data used for the paper. It is easy to read and can be modified for any other data. I will modify it to be more generic for any list files.

When I will have some time, I will also add a pre-trained model on some public data and test subjects to visualise an agent playing.

[crypdick]

@amiralansary ah, ok. So your dataset is not public?

[amiralansary]

The fetal ultrasound data are not public. The MR brain data are public from ADNi datasets http://adni.loni.usc.eduprotocols/
For my experiments, they were pre-processed and reorganised in specific directories. This can be done for any other medical data. I will make these data readers more generic and easier to modify in the near future. Meanwhile I will probably upload a pre-trained model and few sample images from the ADNI database that can be used for playing the agents.

[crypdick]

A handful of pre-processed training example would be greatly appreciated! I just want to get the code up and running so that I can fork it to try to make a Vaa3D plug-in.

[crypdick]

@amiralansary Which one of the ADNi datasets? I've searched through your 2018 paper and Li et al 2018 and I didn't find which one was used. Also, I'd be happy with just a single training example, just to get your code running.

[amiralansary]

@crypdick I just pushed a pre-trained model on the adni data, as well as 10 subjects from the test unseen data. Let me know if you could not manage to get the code running.

[crypdick]

@amiralansary tested, and everything is working!