Code Tests - Presentation stage

[alonshmilo]

Issue: Code Tests - Presentation stage

Explanation:

Code Tests, borderline cases.

Checklist:

• [ ] Code
• [ ] Reading
• [ ] Documenting reading
• [ ] Project book
• [ ] Advisor consult

[stavdv]

Need to get things organized: Deepmedic: Synthetic data, dice and accuracy. Yolo: coco data and our data on several images - high and low resoloution.

[stavdv]

Preliminary checks Deep medic:

• tinyCnn wtih syntetic data.
• path Check in the configuration files.
• Validation of image dimensions in cddHelpers (line 106)
• Usage checks (in ./deepMedicRun line 100)
  • [-newModel] and [-model] options cannot be used together.
  • [-test] cannot be used in conjuction with [-newModel] or [-train].
  • The option [-pretrained] can only be used together with the [-newModel] option:
  • The option [-layers] can only be used together with the [-pretrained] option.
  • The option [-resetOptimizer] can only be used together with the [-train] option.
• Accuracy monitoring per classes for subepoch - (in accurcyMonitor.py)
  • Mean accuracy: (TruePos+TrueNeg)/All predicted voxels.
  • Mean sensitivity: TruePos/RealPos
  • Mean specificity: TrueNeg/RealNeg
  • Mean Dice:[ 2 * (All common between TruePos and RealPos)]/ TruePos + RealPos
• In cnn3d.py:
  • check Mean Validation Accuracy Of Last Epoch And Update Cnns Top Accuracy Achieved If Needed,Called at the end of an epoch, right before increasing self.numberOfEpochsTrained.
  • check Training State Attributes Initialized
• Learning Rate Schedule. (in trainValidate...)
  • check the flag that should have been set True from the start. if training should do Auto-schedule. If we get in here, this is a bug.
  • dimensions check

[alonshmilo]

Yolo checks:

• GT image with original weights
• Scan image with original weights.
• Scan image with our trained weights.

[alonshmilo]

👍