Dataset Augmentation And Balancing

via “multimodal dataset augmentation and transformation”

1.2M image-text pairs with GPT-4V captions.

Unique: Enables systematic augmentation of 1.2M image-caption pairs through deterministic transformations, increasing effective training data size and diversity without requiring additional annotation or API calls

vs others: More efficient than collecting additional images; augmentation strategies are tailored for vision-language tasks (e.g., generating hard negatives) rather than generic image augmentation

via “data augmentation with composition and on-the-fly application”

Unified YOLO framework for detection and segmentation.

Unique: YAML-driven augmentation composition allows non-engineers to modify pipelines without code changes. Mosaic and mixup are implemented as custom ops integrated into the data loader, not post-hoc. Albumentations integration provides 50+ transforms while maintaining YOLO-specific coordinate handling.

vs others: More flexible than TensorFlow's built-in augmentation (YAML config vs code) and more integrated than standalone Albumentations (automatic coordinate transformation for boxes and masks)

via “data augmentation pipeline with geometric and photometric transforms”

OpenMMLab detection toolbox with 300+ models.

Unique: Implements composable augmentation pipelines where transforms are modular components applied sequentially with automatic coordinate transformation for bounding boxes and masks; supports advanced augmentations (mosaic, mixup) that combine multiple images, enabling improved robustness without dataset preprocessing

vs others: More flexible than fixed augmentation strategies because transforms are configurable and composable; more efficient than pre-augmented datasets because augmentation is applied on-the-fly during training; better integrated than external augmentation libraries because coordinate transformation is handled automatically

via “data augmentation with composition and visualization”

Real-time object detection, segmentation, and pose.

Unique: Implements a composable augmentation pipeline with YOLO-specific transforms (mosaic, mixup) and YAML-driven configuration, enabling systematic augmentation experimentation without code changes and with built-in visualization for parameter validation

vs others: More integrated than Albumentations because augmentations are native to the training pipeline, and more specialized than generic augmentation libraries because mosaic and mixup are optimized for object detection

via “intelligent dataset augmentation with version management”

End-to-end computer vision from annotation to deployment.

Unique: Applies augmentation while automatically preserving annotation integrity (bounding boxes, polygons adjusted for transformations), eliminating manual re-annotation; stores augmented versions as separate dataset versions with metadata tracking for A/B testing model performance

vs others: More integrated augmentation than Albumentations (which requires custom Python code) but less flexible than Imgaug for parameter tuning; unique version management allows comparing model performance across augmentation strategies without storage duplication

via “data-augmentation-with-mosaic-and-mixup-strategies”

Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.

Unique: Implements advanced augmentation strategies (Mosaic, MixUp, CutMix) as composable transforms that can be chained and applied probabilistically, with automatic label transformation to match augmented images, rather than simple per-image augmentations

vs others: More sophisticated than Albumentations (which focuses on geometric/color transforms) because it includes Mosaic and MixUp strategies proven effective for YOLO training, and more integrated than standalone augmentation libraries because augmentations are tightly coupled with label transformation

via “data augmentation and filtering for training robustness”

|Free|

Unique: Combines augmentation and filtering in a single pipeline, applying augmentation only to high-quality examples. Uses configurable heuristics for filtering, enabling adaptation to different document types and quality standards.

vs others: More efficient than collecting more training data because augmentation increases diversity; more robust than training on unfiltered data because filtering removes corrupted examples that would degrade performance.

via “multimodal dataset sampling and stratification for balanced model training”

Dataset by mlfoundations. 10,34,415 downloads.

Unique: Enables stratified sampling across document types and content properties at scale, allowing researchers to control training data distribution — most large datasets provide raw access without built-in stratification mechanisms

vs others: More flexible than fixed dataset splits; enables targeted evaluation on specific document categories; supports research on dataset bias and distribution effects

via “dataset curation, augmentation, and preprocessing pipeline”


Unique: Emphasizes data-centric AI philosophy where dataset quality is the primary lever for model improvement, rather than architecture tweaking. Provides systematic approaches to identifying data issues (label noise, distribution shift, class imbalance) and practical augmentation strategies with empirical validation of their impact on model performance.

vs others: More practical and comprehensive than generic data preprocessing tutorials by focusing on deep learning-specific augmentation techniques and providing systematic frameworks for identifying and fixing data quality issues that limit model performance.

via “dataset-augmentation-and-balancing”

via “data augmentation and synthetic sample generation”

via “automated dataset splitting and preprocessing”

via “imbalanced-dataset-rebalancing”

via “automated dataset augmentation and preprocessing”