VisDro_Title is a hybrid version of the VisDrone dataset. The dataset likely contains images for object detection tasks, with 30% of the images being tiled to increase sample diversity. The author, organization, and specific data size are unknown.
Use Cases
- Train object detection models based on the described drone imagery.
- Benchmark model performance on a hybrid dataset containing both standard and tiled images.
- Study the effect of data augmentation via tiling on model robustness for aerial scenes.
Strengths
- Builds upon the established VisDrone benchmark dataset for drone-based object detection.
- Introduces a specific data variation with 30% of images being tiled, which may aid in model generalization.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- Kaggle
- Collection Method
- Derived from the VisDrone dataset, with a portion of images processed via tiling.