datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
VisAssistdefectforge-visa-synthetic
DefectForge VisA Synthetic Defects
Synthetic defect images and generation-time masks for the VisA pcb1 and capsules
objects. Each object is generated from only 10 real anomalous training images, while the
frozen high-shot test partition is never visible to generation, filtering, or quality
reference sets.
繁中摘要：這是 VisA pcb1／capsules 的少樣本工業瑕疵合成資料。每個物件只用
10 張真實瑕疵訓練圖；mask 是生成時使用的標註，不是模型事後預測。資料同時提供
filtered 與 unfiltered 版本，並公開 provenance 與 test SHA-256 blocklist。
What is… See the full description on the dataset page: https://huggingface.co/datasets/steven0226/defectforge-visa-synthetic.VIS-APP-Bench
anchor_tasks_web — Dataset
This dataset is part of the benchmark presented in the paper VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents.
Project Page | GitHub Repository | Paper
A web-app generation benchmark. Each task is a multi-page UI taken from a public Figma community file. For every task we ship the textual page descriptions, the rendered mockup PNGs, the Figma node structure, the per-page click-annotations, and the distilled data-testid "anchors"… See the full description on the dataset page: https://huggingface.co/datasets/JunJiaGuo/VIS-APP-Bench.