Datasetpublic

kfkas/streamingvlm-task-aware-vqa-suite-rowwise

StreamingVLM Task-Aware VQA Suite (Row-wise) Viewer-ready evaluation samples for task-aware visual-token sensitivity experiments. Every config contains 3,000 deterministic manifest-order samples with the image embedded in each row. Task groups task_group Dataset configs coarse_object_presence pope, repope, hpope general_scene_understanding vqav2, gqa fine_grained_visual_evidence gqa_attribute, mmbench textual_fine_grained_evidence textvqa, docvqa… See the full description on the dataset page: https://huggingface.co/datasets/kfkas/streamingvlm-task-aware-vqa-suite-rowwise.

sourceHugging Faceupdated 2mo agoView on Hugging Face

0likes191downloads

Dataset Card

StreamingVLM Task-Aware VQA Suite (Row-wise)

Viewer-ready evaluation samples for task-aware visual-token sensitivity experiments. Every config contains 3,000 deterministic manifest-order samples with the image embedded in each row.

Task groups

task_group	Dataset configs
`coarse_object_presence`	`pope`, `repope`, `hpope`
`general_scene_understanding`	`vqav2`, `gqa`
`fine_grained_visual_evidence`	`gqa_attribute`, `mmbench`
`textual_fine_grained_evidence`	`textvqa`, `docvqa`

Columns

—image: image payload rendered by the Hugging Face Viewer
—dataset_key: source dataset identifier
—task_group: research task group used for analysis
—category: dataset-specific subtype such as POPE adversarial
—question, answer, all_answers: evaluation input and references
—metadata_json: complete original manifest row

gqa_attribute is a derived attribute-focused subset rather than an official GQA split. MMBench remains a broad benchmark; use its category metadata when selecting fine-grained samples.

The complete manifests and deduplicated image archives remain available in streamingvlm-task-aware-vqa-suite-direct.