osc.train
Main training methods.
Functions
Compute number of batches in one test epoch. |
|
Compute number of batches in one training epoch. |
|
Compute number of batches in one validation epoch. |
|
Build test dataset. |
|
Build training dataset. |
|
Build validation dataset. |
|
Build VQA dataset for linear probing. |
|
|
|
Build global loss function. |
|
Build object loss function. |
|
Build model |
|
Build optimizer for training. |
|
Build learning rate scheduler for training. |
|
Prepare a batch of images for visualization. |
|
Log model parameters as a table. |
|
|
|
Run one epoch of training. |
|
Run train, val, viz for a certain number of epochs. |
|
Run one validation epoch. |
|
Run inference on a single batch of images and visualize everything! |
|
Classes
Function type signature for model loss functions. |