Closed QJvic closed 5 years ago
And my pytorch is 0.4 , does this have influence?
Hi @QJvic
You should downgrade to pytorch 0.3.1
but it seems that there is something wrong with the paths.
What are your data paths in the neptune.yaml
?
Your masks directory is probably named 'masks_overlayed_eroded_0_dilated_2', whereas the model is looking for 'masks_overlayed_eroded_0_dilated_0'.
you need to go into src/pipeline_config and edit line 13 to say Y_COLUMNS = ['file_path_mask_eroded_0_dilated_2']
@QJvic @carbonox-infernox
return joblib.load(filepath) File "C:\Users\MSIK\Anaconda3\lib\site-packages\sklearn\externals\joblib\numpy_pickle.py", line 590, in load with open(filename, 'rb') as f: FileNotFoundError: [Errno 2] No such file or directory: 'D:\\crowdAI\\test0118\\masks_overlayed_eroded_0_dilated_2\\train\\masks\\000000142365'
I had the same problem. My platform is : Win10+anaconda, Python3.7 and Pytorch1.0.
The following code executes correctly.
python main.py -- prepare_masks python main.py -- prepare_metadata --train_data --valid_data --test_data
It seems that my path settings are correct, but there is no file suffix. Like"000000142365.png"
I hope you can help me. Thanks!
Hi there @xdsonglinliu.
Can you check that this file exists and you can open it separately? For example go to the notebook and run
from sklearn.externals import joblib
joblib.load(TYPE_FILEPATH)
It is also possible, that there are some assumptions about path parsing that we made because we developed the codebase entirely on linux.
@jakubczakon Thanks a lot. With your help, the problem has been solved.
@xdsonglinliu @jakubczakon
Hi, I am also encountering his problem as well. However, I noticed that the file 000000142365.png
does not exist in the masks
directory after generating the masks through python main.py -- prepare_masks
. The difference is I used annotation-small.json
instead of annotation.json
, since my machine couldn't handle the large amount of files, similar to the situation in #166. This might be the cause why the particular 000000142365.png
file was not created.
Is it necessary to use the larger annotation.json
file to proceed with training? Thank you very much.
@cvKDean
I am afraid something went wrong and you need to rerun with the full annotation.json
file.
This file is simply not a part of the annotation-small.json
.
@jakubczakon
Thank you for your response. I ran python main.py -- prepare_masks
again, this time with annotation.json
. And I confirmed that the file 000000142365.png
was present in the directory. However, I still encounter the same problem. It seems that loaders.py
has the wrong file path (no .png suffix).
I also checked the generated metadata . csv file via python main.py -- prepare_metadata --train_data --valid_data --test_data
and I also confirmed that the filepaths are correct, and they have .png suffixes. Am I missing something? Thank you for your patience.
Edit: I am trying to see where the code reads the stage1_metadata.csv
file and passes the filepath of the image to load_image
in loaders.py
, but I can't see exactly where img_filepath
is being extracted from.
@xdsonglinliu
Can you explain how to fix the problem Please!
@jakubczakon can you please tell about this error: FileNotFoundError: [Errno 2] No such file or directory: 'data/meta_data/masks_overlayed_eroded_0_dilated_0/train/masks/000000153612.png'
I have already prepared masks from python main.py prepare_masks python main.py prepare_metadata --train_data --valid_data and has generated a metadata.csv file
I use this in WIN10 ,anaconda. When I typed
python main.py -- train --pipeline_name unet_weighted
,I got errors as follows.It seems like can't find the png file.But I search it in folder and find 000000142365.png does actually exist. Wish for your replay.