Setcon Training Datasets: Annotations for Open-Ended Referring Segmentation

SetCon training datasets provide the annotations used for training and evaluating the SetCon model for open-ended referring image and video segmentation. The dataset was created by author rookiexiong and was last updated on May 20, 2026. Its structure includes separate directories for image and video annotations, with files named for specific subsets like 'grefcoco' and 'muse'.

Use Cases

• Training models for referring image segmentation based on the provided image annotations.
• Evaluating model performance on open-ended segmentation tasks using the evaluation annotations.
• Developing video segmentation models based on the described video annotation structure.
• Benchmarking progress in open-ended visual concept prediction as described in the associated paper.

Strengths

• Annotations are specifically designed for the novel 'Set-Level Concept Prediction' task described in the associated paper.
• Dataset structure is organized into dedicated image and video directories, suggesting clear separation of modalities.
• Includes annotations from multiple sources, such as 'grefcoco' and 'muse', indicating a composite training set.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and sample data are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

rookiexiong on Hugging Face

Collection Method

Annotations used for training the SetCon model, as described in the associated paper.

Time Range

null

Freshness

Last updated 2026-05-20 03:40:59; freshness should be verified.

Geography

null