Parameters to keep track for trainings

[Rob174]

Data:

• range of images used
• images excluded ⚠️ more raw than preprocessed
• resolution distribution
• number of source images used
• classes available and mappings
• distribution of max/min length of oil discharges
• type de valeurs prédites Preprocessing:
• Grid / patchsize
• number of patches
• Preprocessed images or not
  • which preprocessing pipeline (parse it to json)
• data augmentations (with parameters) AI:
• loss
• metrics
• optimizer
• model
• confusion matrix

Objects used:

• make an id to log the modules used for the augmentations, the model.... Write in the logs the id and the name. Change the id for each version of the code: use the commit hash
• log the commit hash

Stat on number of pixels classified

[Rob174]

As a first step a quick iterative search to understand the which parameters are key to achieve best performances. We will make a training, keep it only if it produces interesting results and test another set of parameters.

[Rob174]

As a second step, we will vizualize results with an RGB image where the probability of each class is mapped to one channel (only possible for the first dataset with 3 classes)

[Rob174]

We will use early stopping to cut training time We will make multiple runs with the same set of parameters and observe the mean accuracy/loss

[Rob174]

The results will be presented in a jupyter notebook

When the project became stable, a high level api would be better to ease the work of future students