AniMINT: UI Animation Interpretation Dataset

Dataset Description

AniMINT is a dataset for evaluating whether vision language models (VLMs) can understand UI animations beyond static screenshots. The dataset contains 300 densely annotated UI animation videos from mobile, web, and desktop interfaces. Each animation is annotated with:

1. Start and end frame of the animation
2. Animation region of interest(s)
3. Context information
4. User input information, if any
5. 10 unique human-annotated, open-ended descriptions of the animation effect
6. 10 unique human-annotated, open-ended descriptions of the animation meaning
7. Categorization of animation purpose

The dataset is intended to support research on:

• UI animation understanding;
• multimodal and video-language evaluation;
• UI agent perception;
• motion-grounded interface reasoning.

This dataset accompanies the paper:

Beyond Screenshots: Evaluating VLMs’ Understanding of UI Animations

If you use AniMINT, please cite:

@misc{liang2026beyondscreenshots,
  title         = {Beyond Screenshots: Evaluating VLMs' Understanding of UI Animations},
  author        = {Liang, Chen and Jiang, Xirui and Deng, Naihao and Adar, Eytan and Guo, Anhong},
  year          = {2026},
  eprint        = {2604.26148},
  archivePrefix = {arXiv},
  primaryClass  = {cs.HC},
  url           = {https://arxiv.org/abs/2604.26148}
}

Dataset Summary

Property	Value
Dataset name	AniMINT
Full name	UI AniMation INTerpretation Dataset
Number of clips	300
Modality	UI animation video + text annotations
Source platforms	Mobile, web, desktop
Main platform distribution	Mobile: 75.00%; Web: 15.67%; Desktop: 9.33%
Median animation duration	3.59 seconds
Annotation language	English
Video/interface language	Primarily English
Annotator region	United States
License	CC BY-NC-ND 4.0

Animation Purpose Labels

AniMINT uses seven purpose categories for UI animations:

Label	Description
Transition	Animations that support layout or state changes in the interface.
Demonstration	Animations that reveal or explain the behavior, functionality, or structure of the interface or its elements.
Guidance	Animations that guide the user toward an intended interaction.
Feedback	Animations that provide visual responses to user interactions.
Visualization	Animations that represent system status, data, progress, or other information.
Highlight	Animations that emphasize specific content or draw the user’s attention to key elements.
Aesthetic	Animations that enhance visual appeal, create emotional impact, or improve user experience without primarily conveying required information.

Meaning Interpretation and Effect Description Annotations

AniMINT includes open-ended human text annotations for each UI animation video. Each video contains:

Annotation Type	Count per Video	Description
animation_meaning_annotations	10	Human-written descriptions of what the animation communicates or means in the UI context.
animation_effect_annotations	10	Human-written descriptions of the visible animation effect, motion, or perceptual change.

These annotations support evaluation of whether a model can go beyond detecting motion and produce interpretations that align with human understanding. The meaning annotations focus on the interpretation of the animation, such as whether it signals an error or system state., or attract users' attention. The effect annotations focus on the visible motion or visual transformation, such as an element shaking, fading, filling, bouncing, moving, changing color, or resizing.

For example, a shake animation that indicates a failed login attempt may have an animation effect description of: "The login field page becomes outlined in red and rapidly shakes 3 times when the user taps the sign in button." and an interpretation of: "The animation is telling the user that signing in was unable to be completed and that some error has occurred that has prevented the sign in."

DMCA Notices and Data Removal Requests

AniMINT is released for non-commercial research and evaluation purposes. The dataset may include recordings of third-party user interfaces, logos, or other UI assets. The research team has cut the video to primarily focus on the research evaluation purposes. The collection source for each animation is attributed in the datasource.csv. All third-party rights remain with their respective owners.

If you are a rights holder, platform operator, developer, or individual who believes that a dataset item should be removed or reviewed, please contact us at researchpubacc at gmail.com with details including the dataset item ID or filename, the reason for the request, your relationship to the content or rights holder, and a contact method for follow-up.