DOCUMENTATION.md

API Overview

Modules

• data
• data.augment
• data.collate
• data.data_factory
• data.dataset
• data.transforms
• eval
• hp_optim
• models
• models.backbone_factory
• models.byol
• models.byol.model
• models.mae
• models.mae.lr_sched
• models.mae.model
• models.mae.pos_embed
• models.model_factory
• models.simclr
• models.simclr.encoder
• models.simclr.head
• models.simclr.loss
• models.simclr.model
• pipeline
• pipeline.callback_factory
• pipeline.lightning
• train
• utils
• utils.image
• visualize

Classes

• augment.Dilation
• augment.Erosion
• augment.GaussianNoise
• augment.PairTransform
• dataset.ICDARDataset
• model.BYOL
• model.EMA
• model.NetWrapper
• lr_sched.CustomScheduler
• model.MAE: Masked Autoencoder with VisionTransformer backbone
• encoder.ResNet50Encoder
• head.ProjectionHead
• loss.ContrastiveLoss
• model.SimCLR
• lightning.LightningPipeline

Functions

• collate.collate_factory: Custom collate function for each model.
• data_factory.data_factory: Data loader factory based on dataset name.
• transforms.transform_factory: Transform factory for self-supervised models
• eval.execute: Evaluation entry point.
• hp_optim.objective: Objective function for Optuna.
• backbone_factory.backbone_factory: Backbone factory for self-supervised models
• model.MLP: Simple MLP with ReLU activation and batch norm
• model.SimSiamMLP: SimSiam MLP with ReLU activation and batch norm
• model.default
• model.flatten
• model.get_module_device
• model.loss_fn: Negative cosine similarity loss as defined in the paper
• model.set_requires_grad
• model.singleton: Singleton pattern decorator
• model.update_moving_average
• pos_embed.get_1d_sincos_pos_embed_from_grid: embed_dim: output dimension for each position
• pos_embed.get_2d_sincos_pos_embed: grid_size: int of the grid height and width
• pos_embed.get_2d_sincos_pos_embed_from_grid
• pos_embed.interpolate_pos_embed
• model_factory.model_factory: Model factory for self-supervised models
• encoder.test
• loss.test
• callback_factory.callback_factory: Model callback factory
• train.execute: Configuration based model training entry point.
• image.img_is_color: Check if an image is color or grayscale.
• image.show_image_list: Shows a grid of images, where each image is a Numpy array. The images can be either
• visualize.plot_features: Plot embeddings. This is a wrapper around : func : tsne. TSNE to make it easier to visualize the model's performance.