Progress UNet July Oskar

What I did:

1st push: "UNet class implementation, TODO" Check if: Augmentation -> image resize shape is power of 2? Data -> original images different sizes? square? power of 2?

function: model_summary-> summary (model size, depth, etc)

class: qim_UNet -> choose between small, medium, large dataset

---

2nd push: "working on hyperparameter class" class: qim_hyperparameters -> insert model,epochs,lr,optimizer, momentum, weight decay, loss returns a 'hyper_dict' with {optimizer,criterion,n_epochs}

---

3rd push: "hyperparameter class implementation + TODO models" TODO: qim3d/utils/models.py 'preds > 0.5' -> outputs of model not in [0,1], need to use sigmoid beforehand

class: qim_hyperparameters: ordering for better understanding (_optimizer and _loss_functions)

---

4th push: "implementing training function" function: train_model -> takes model, qim_hyperparameters, trainloader,valoader. performs training loop. returns 2 dict: train_loss and val_loss each dict has: loss per epoch, and loss per batch.

minor changes to: class: qim_hyperparameters

---

5th push: "loss visualization function,creation condensed notebook" created new jupyter_notebook with condensed version

function: visualizations.py -> plt_metrics function: plots output of train_model

---

6th push: "Solution tqdm training + update both notebooks" added tqdm loop to train_model function (/!\ see issue)

---

7th push: "minor changes ValueError, log.warnings" Changed some ValueError messages from previous functions made: changed to: "invalid ___: {input}, please use: {suggestions}"

---

8th push: "removed batch" removed small test made in previous push, which wasn't removed.

requested review from @fima, @s184364, and @s184058

assigned to @s184364

added 1 commit

• 4523e8e0 - Resolved Felipe's comments for now, however more changes needed

Compare with previous version

added 1 commit

• 0d9ea53f - Unet Model is now nn.Module

Compare with previous version

added 2 commits

• 2f1b21a7 - Default to no resizing. If data is resized, log.info. Problems with resizing to non-square images.
• a9b90799 - Fixed Resizing issue. User chooses whether to resize, and UNet runs for both...

Compare with previous version

added 6 commits

• f597d896 - Function for loading h5 files
• d4555e70 - Data explorer with basic viz
• 6ac32b6b - Quick fix to Data explorer
• ecbb72f9 - Changing default path for Data Explorer
• 5686e692 - First version of tiff stack loader
• d58fa3ae - data_set/loaders implemented in function. plotting moved to training function.

Compare with previous version

resolved all threads

approved this merge request

unapproved this merge request

approved this merge request

added 1 commit

• 7f19e242 - Removed long notebook for Unet

Compare with previous version

added 16 commits

• 7f19e242...2d546cd3 - 15 commits from branch main
• adba82bf - merge with main

Compare with previous version

approved this merge request

mentioned in commit 3ff594d9

merged