COMP700 Year Long Project: Deep Learning for Tracking Moving Cells in Time-lapse Video Sequences
This project explores techniques to segment 10 datasets from the Cell Tracking Challenge. After considering many traditional segmentation techniques, the project shifts focus to Neural Network segmentations via a U-Net architecture.
To use the notebooks, the repository may be pulled into a desirable folder. The folder path that the author used was: C:\Users\...\Documents\Github\COMP700
In addition to that folder, the extracted datasets were loaded into the directory: C:\Users\...\Documents (Notice how it is 2 steps behind the Github repository - to avoid commits upwards of 1 GB a push)
The files used take up a lot of space, as just the zipped folder containing the training and test sets for the 10 types of cells, takes up ~7GB of file space
For this reason, the author suggests downloading the source files (which all lie in 1 zipped folder) and then doing the following:
- In C:\Users\...\Documents, create a folder called "Comp700_DataSets"
- Inside that folder, C:\Users\...\Documents\Comp700_DataSets, extract the contents of the source file. This should yield 2 files inside C:\Users\...\Documents\Comp700_DataSets: C:\Users\...\Documents\Comp700_DataSets\Extracted C:\Users\...\Documents\Comp700_DataSets\OriginalZipped
- For each of the 20 datasets inside C:\Users\...\Documents\Comp700_DataSets\OriginalZipped, extract the contents into the C:\Users\...\Documents\Comp700_DataSets\Extracted path. This should not setup the dependencies for the project.
- With reference to notebook 001, any additional folders will be generated by the notebooks, so the most important initial folder setup is described above
Please note that there are some notebooks that had to be created and used in Google Colab. These files are inside a folder called "Google Drive and Google Colab Content" in the Github repository of this project. Those folders should be loaded into the appropriate Google Drive directory, and the appropriate folders generated offline should be uploaded, described as:
drive > MyDrive > COMP700_Images > COMP700_Processed_Training_GT
drive > MyDrive > COMP700_Images > COMP700_Processed_Training_ST
drive > MyDrive > COMP700_Images > COMP700_Raw_Training_GT
drive > MyDrive > COMP700_Images > COMP700_Raw_Training_ST
Those 4 folders are generate offline, by notebook 010
Uploaded on the UKZN website is the presentation material for this project.
For any questions, please feel free to reach out to the author, via his Github account: 'alexgoudemond'