Collab not recognizing labeled data (.h5 file)

I’ve uploaded my project to google drive, and am interested in running it on Google collab. I’ve successfully linked my path_config to my config.yaml file but now when I run the check_labels method it’s says I do ‘not appear to have labeled data!’ in the folders (11 total). All folders have their respective csv and h5 files and when I run this method on my desktop, it produces the correct output with subdirectory folders with labeled frames.

When I run the create_training_dataset method, it says that it can’t specifically find the .h5 files. I can clearly see the .CSV files have were updated when I was labeling via GUI, and I suppose I just assumed the .h5 file (I can also see that when the .csv is modified, the .h5 is also modified).

Anyone have any ideas/suggestions of what is going on here, and how to proceed?

It is likely that you have a package that is not correct; i.e. Colab updates very often, whereas we have some fixed dependencies. I checked colab today, and updated the notebooks. Please use them: i.e. look at the expected outputs from the DEMO, and then use the one that is designed for your own project…

and

Hi MWMathis,

Thanks for you quick response!

So after attempting to re-run with my demo_your_own_data Collab with the new updates from the other demos, I kept running into more or less the same problem.

(1) Check the labels - same error
“path of designated folders Doesn’t appear to have labeled data!”
(2) Create a Training Dataset - new error (instead of saying it either can’t find the .h5 file or the .h5 file isn’t labeled) it says:
" File “< ipython-input-6-7258774aaa0d >”, line 9"
SyntaxError: invalid syntax.


So I just tried running the updated Colab_TrainNetwork_VideoAnalysis demo and it’s still having a problem reading the .h5 file.

(1) Check the labels
“Creating images with labels by Mackenzie.
Attention: /content/drive/My Drive/DeepLabCut/examples/Reaching-Mackenzie-2018-08-30/labeled-data/WILL BE AUTOMATICALLY UPDATED BY DEMO CODE does not appear to have labeled data!
If all the labels are ok, then use the function ‘create_training_dataset’ to create the training dataset!”

(2) Create a Training Dataset - new error - it actually creates the training dataset (which makes me think mydemoCollab wasn’t fully updated currently) but it still isn’t reading the .h5 file.
“Creating images with labels by Mackenzie.
Attention: /content/drive/My Drive/DeepLabCut/examples/Reaching-Mackenzie-2018-08-30/labeled-data/WILL BE AUTOMATICALLY UPDATED BY DEMO CODE does not appear to have labeled data!
If all the labels are ok, then use the function ‘create_training_dataset’ to create the training dataset!
/content/drive/My Drive/DeepLabCut/examples/Reaching-Mackenzie-2018-08-30/labeled-data/WILL BE AUTOMATICALLY UPDATED BY DEMO CODE/CollectedData_Mackenzie.h5 not found (perhaps not annotated)
Annotation data was not found by splitting video paths (from config[‘video_sets’]). An alternative route is taken…
The following folders were found: [‘reachingvideo1’]
Downloading a ImageNet-pretrained model from http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz
The training dataset is successfully created. Use the function ‘train_network’ to start training. Happy training!”

yeah for the demo data, you would need to update the config file,…ti the correct paths

Okay thanks, that was correct, and I was able to get it running.

I’ve trained a network to 205,000 iterations, and then it stopped training, so I ran it again starting at that checkpoint. Starting again, renewed the iteration sequence back to 0, so then the second time around ran it to 230,000 iterations. Technically would the amount of iterations run on this be the sum of those two then?

** The question below was posted to github: https://github.com/AlexEMG/DeepLabCut/issues/352

Also I don’t know if this might be a new question to post, but when I run evaluation results, it runs without error, and supposedly creates all the correct folders with three files (results.csv (153bytes), results.h5(1KB), .h5 (215KB)) But they seem to be empty files and there is no plotting folder (even though I set it to True). And no reported evaluation results.

** projectname/shuffle1_230000 with # of trainingiterations: 230000

This net has already been evaluated!

The network is evaluated and the results are stored in the subdirectory ‘evaluation_results’.

If it generalizes well, choose the best model for prediction and update the config file with the appropriate index for the ‘snapshotindex’.

Use the function ‘analyze_video’ to make predictions on new videos.

Otherwise consider retraining the network (see DeepLabCut workflow Fig 2)**

Is it not reading the .mat file correctly?
(Also should I post this as a separate post?)

Thanks!

Fixed on Github: https://github.com/AlexEMG/DeepLabCut/issues/352