Sign in to view source links and access this dataset
Description
SPRIGHT (SPatially RIGHT) is a large-scale vision-language dataset created by re-captioning approximately 6 million images from four widely-used datasets: CC12M, Segment Anything, COCO Validation, and LAION Aesthetics. The dataset was uploaded by SPRIGHT-T2I and was last updated on October 9, III. This repository contains the re-captioned data from CC12M and Segment Anything, while COCO data is linked separately; images from LAION are not released as the parent images are private.
Use Cases
Training vision-language models for spatial understanding based on the dataset's spatially focused captions.
Benchmarking image-text alignment models using the re-captioned data from multiple large-scale sources.
Fine-tuning models for tasks requiring spatial reasoning in visual scenes, as suggested by the dataset's description.
Researching the impact of spatial captions on multimodal AI performance.
Strengths
Built from approximately 6 million images, indicating a large scale.
Aggregates and re-captions images from four established datasets: CC12M, Segment Anything, COCO Validation, and LAION Aesthetics.
Focuses on spatial reasoning, a specific and potentially valuable niche for vision-language tasks.
Limitations
Row count, file formats, and column-level documentation are unknown, which may limit suitability assessment.
Images from the LAION source are not released, reducing the completeness of the aggregated collection.
License information is unknown, which may restrict usage.