Topic Recognition #438
Open
Topic Recognition #438
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
To accept changes made in most recent pull request
All files listed in the report spec sheet have been created and README instructions from spec sheet have been added to the README file.
Updated to prevent local data files from being uploaded
Due to errors in data loading, file name has been changes (will revert to more appropriate name at end of project)
Updated and debugged to sample images for display
Note that train code references module which is not yet successful and thus not committed. Code adapted from https://github.com/keras-team/keras-io/blob/master/examples/generative/vq_vae.py which is linked to in paper 11 from assignment resources
Successfully load in and display image and mask pairs
Tweaking source code to work for black and white data instead of rgb values in a .cvs file
Note to marker: I am about to do a bunch of commits in a short period of time - this is because I've been working in google colab & forgot to do commits. I've split these into where I would have committed had I remembered to do that properly - each point where I copied across to my personal device to save (using undo, this is the real code I had at these times). Please don't be too harsh.
Convolution and deconvolution blocks built
Unet created from convolutional blocks with snapshots
Training code written - currently throwing memory errors
Unnecessary code removed to help with debugging
Incorrect input shape found to be causing errors
Images were too large & causing memory errors. Also, Epochs were taking too long and reached a reasonable val-accuracy after only 10 epochs.
Source code merged training and testing data - this would invalidate results - fixed.
The code now runs and produces fairly decent labels for the images (when run to 30 epochs, the labels almost seem better than the true labels for identifying skin discolourations - possibly over trained, so 10 epochs will be kept as optimal)
The labels were being printed in a colour map - updated to show correct 'grey scale' black and white
Updated loss to use dice coefficient and made improvements model fit call.
Put the code from the previous commit in the wrong line - fixed
Correct dice similarity and loss implemented. Epochs raised from 10 to 15 and learning rate specialized to 0.00003 to regain good accuracy results. Evaluation and prediction code added to verify functionality. Prediction now outputs 5 comparisons instead of 1.
Added some general comments & produced ReadMe document. Also updated File Name
|
This is an initial inspection, no action is required at this point
|
Good Practice (Design/Commenting, TF/Torch Usage)Adequate use and implementation Recognition ProblemSolves problem Commit LogMeaningful commit messages, could be more descriptive -1 DocumentationReadMe acceptable/good Pull RequestSuccessful Pull Request (Working Algorithm Delivered on Time in Correct Branch) |
|
Feedback incorporated - no conflicts remaining |
|
Looks like the changes aren't showing up, I still see the same issues in the files changed tab. Can you check if these changes have indeed been reverted? (Will award feedback marks because fix was attempted, but can't merge until fixed) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Mikayla Staples (46413587) COMP3710 Final Project:
Improved U-Net (task 1) applied to the 2017 International Skin Imaging Collaboration (ISIC) Dataset with Segmentation, attaining Sørensen–Dice similarity coefficient of 0.926 on the test set after 15 Epochs of training.