README.md

metadata

pretty_name: WildGUI Screenshots (part16–19)
license: cc-by-nc-4.0
language:
  - en
tags:
  - gui-agents
  - gui-grounding
  - interaction-trajectories
  - video2gui
  - wildgui
  - screenshots
  - images
  - web
  - desktop
  - mobile
size_categories:
  - 1M<n<10M

WildGUI Screenshots (part16–19)

This repository hosts the screenshot images for part16–part19 of WildGUI, the dataset introduced by Video2GUI. It extends the main release at xwm/WildGUI, which already contains all annotations plus the screenshots for part1–part15.

The two repositories are split as follows:

Repository	Annotations	Screenshots
xwm/WildGUI	All parts (JSONL)	part1–part15
joker-112/WildGUI_Screenshots (this repo)	—	part16–part19

So the annotations for part16–part19 live in xwm/WildGUI, while their screenshot frames live here. To work with part16–part19 you need both: the trajectories from xwm/WildGUI tell you what action happens and where, and each action points at one screenshot frame stored in this repo.

File Layout

Screenshots are grouped by the same partN shards used for the annotations, then packed into uncompressed tar archives (the frames are already JPEG-compressed):

screenshots/
  part16/
    wildgui_part16_images_000001.tar
    wildgui_part16_images_000002.tar
    ...
  part17/
    wildgui_part17_images_000001.tar
    ...
  part18/
    ...
  part19/
    ...

Each tar holds up to ~20,000 frames. Inside a tar, every frame is stored under a per-video directory:

{video_id}/screenshot_{MM_SS}.jpg

{MM_SS} is the action timestamp normalized to zero-padded minutes_seconds (e.g. the annotation timestamp "00:18" → screenshot_00_18.jpg).

Linking an Annotation to its Screenshot

Each trajectory action in xwm/WildGUI (for part16–part19) maps to exactly one frame in this repo. Given a record's video_id and an action's timestamp:

1. Take the partN matching the annotation shard (e.g. wildgui_part17.jsonl → screenshots/part17/).
2. Normalize the timestamp to MM_SS: timestamps like "00:18", "1:05", or a raw second count are converted to total minutes_seconds, each part zero-padded to two digits.
3. The frame is {video_id}/screenshot_{MM_SS}.jpg, found inside one of that part's wildgui_part{N}_images_*.tar shards.

def timestamp_to_suffix(ts: str) -> str:
    """'00:18' -> '00_18', '1:05' -> '01_05', '78' -> '01_18'."""
    ts = str(ts).strip()
    if ":" in ts:
        total = 0
        for part in ts.split(":"):
            total = total * 60 + int(part)
    else:
        total = int(float(ts))
    minutes, seconds = divmod(total, 60)
    return f"{minutes:02d}_{seconds:02d}"

# For annotation record {"video_id": "tPPzO7wot9s", ...}
# and action {"timestamp": "00:18", ...}:
#   arcname = "tPPzO7wot9s/screenshot_00_18.jpg"

Downloading

Download a single part with the huggingface_hub CLI:

hf download joker-112/WildGUI_Screenshots \
  --repo-type dataset \
  --include "screenshots/part16/*" \
  --local-dir ./wildgui_screenshots

Then unpack the tar shards (each expands into {video_id}/screenshot_*.jpg):

for t in ./wildgui_screenshots/screenshots/part16/*.tar; do
  tar -xf "$t" -C ./wildgui_frames
done

For streaming pipelines (e.g. WebDataset), each tar can also be read directly without extracting to disk.

Notes

• This repo covers only part16–part19. For part1–part15 screenshots and for all annotations, use xwm/WildGUI.
• A small fraction of annotated frames may be missing from the packed shards (source frame unavailable at pack time); treat a missing {video_id}/screenshot_{MM_SS}.jpg as a skippable example rather than an error.
• Frames are derived automatically from tutorial videos and may contain noise. Validate for your own downstream training or evaluation setting.

Intended Use

These screenshots are intended for research on GUI agents, GUI grounding, action prediction, interaction trajectory modeling, and multimodal agent pretraining, paired with the annotations in xwm/WildGUI.

Citation

If you use these screenshots, please cite the Video2GUI paper:

@misc{xiong2026video2gui,
  title = {Video2GUI: Synthesizing Large-Scale Interaction Trajectories for Generalized GUI Agent Pretraining},
  author = {Xiong, Weimin and Gu, Shuhao and Ye, Bowen and Yue, Zihao and Li, Lei and Song, Feifan and Li, Sujian and Tian, Hao},
  year = {2026},
  eprint = {2605.14747},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  doi = {10.48550/arXiv.2605.14747},
  url = {https://arxiv.org/abs/2605.14747}
}