Iteration 2 DMC

Iteration goals:

In this iteration we will make a progress in the project. After making DeepMedic work, and only reading about Cascaded-FCN for RCF and Faster-RNNLM for classifier, we will dive into those 2 projects, understanding what's going on there. Plus, taking what we need - RCF and classifier. Please review the important links found downstairs and see all the work we've done.

Lists of goals:

• Improve DeepMedic run.
• Classifying by yolo

use cases:

• Learning.
• Classifying.

Retro prospective - goals and results:

• To be written in the end of iteration

Roles:

Team: Alon and stav
Generally:

• Alon - Classifier
• Stav - DeepMedic

Issue Management:

Issue Board

Iteration 1 Issues

Tests:

• According to tests plan as shown on report - functional and non-functional tests, being documented.
• Checks for Cascaded and rnnlm.

Points Summary:

• Faster-RNNLM code for classifier - 13 points
• Cascaded-FCN code - CRF method - 13 points
• DeepMedic code - 13 points
• Yolo classifier - 13 points
• Code Tests - 3 points
• Code review - 3 points
• Report - 2 points
  Total: 60 points

DMC - DeepMedicand Classifier

This stage deals with those 2 main subjects that will construct our project. Those issues will be reviewed deeply. Every element that we will need to take to our project will be taken and implemented as an initial state, towards 85% finish of the project. On the one hand - we want to modify DeepMedic into our needs deeply, and making the classifier work.

Important Links

• Working Document - All information gathered through this iteration.
• Research Blog

Iteration Summary:

• As a summary, we have started to work on real data in DeepMedic. It now works on real ab scans, showing segmentations on the ribs.
• Yolo is almost transformed to our data, but it need better training - it takes time. The coordinates files were converted by python code to darknet's format. Now the training process should run.

To improve:

• We have to concentrate on the important things and start closing big and small issues towards the finish.
• Working in the right order, with proper filing and documenting in dropbox, git and personal disks.
• Working together and alone. *Synthetic data came, and it made it much easier. If it got to us sooner, we were was ahead.

To sum it up:

• Now entering the final stage of the project - until June 19 2017, all should be perfect.
• It means that we have to get on the high way and continue work hard.
• The small details is important. Maybe we have to start with big things, but we have to pay attention to the small details, because money-time is here, and there will not be another time.

Next Iteration:

Iteration 3 LCP