Data Input & Preprocessing

Components

ImageBaseDatasetHandler

Serves as the abstract base for handling single-frame image datasets, defining the interface and common logic for loading, standardizing, and preparing individual image frames for subsequent processing.

Referenced Source Code

SpecificDatasetImplementations

Concrete dataset classes that inherit from or utilize `ImageBaseDatasetHandler`, providing dataset-specific logic for parsing annotations and locating files for various image datasets.

Referenced Source Code

DataAugmentationModule

Provides a comprehensive set of functions for augmenting image and pose data, crucial for increasing training data diversity and model robustness by applying various transformations.

Referenced Source Code

ROMPGeneralUtilities

Offers general utility functions for the ROMP project, including fundamental image preprocessing, temporal smoothing, and extracting frame paths, which are leveraged across different data handling components.

Referenced Source Code

Trace2VideoDataLoader

Specializes in loading and preparing video sequences specifically for the trace2 model, handling the complexities of video input and ensuring readiness for sequential model processing.

Referenced Source Code