Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
image
imagewidth (px)
1.08k
1.8k
label
class label
2 classes
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
01b41279a9fef4a1080e3fad9b3be9ac8
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816
11b9be97320alled5a6183455ae92a816

HyperTrack Preview Subset Data Card

Dataset Summary

This repository contains a filtered preview subset of HyperTrack for data-format inspection and offline mobile GUI-agent evaluation. Each episode represents a static interaction trajectory collected from Chinese Android applications. The packaged format includes a high-level user instruction, ordered screenshots, textual screen observations, normalized structured action annotations, spatial grounding annotations where applicable, and device/episode metadata.

The preview subset is intended to show the release format and support local evaluation-tool development.

Files

  • test_filtered/: filtered preview episodes.
  • filter_hypertrack_episodes.py: script used to normalize actions and generate test_filtered/.
  • episode_categorization.json: source split membership file used to select test episodes from the local archives.

Episode Format

Each episode is a directory named by episode_id:

data/test_filtered/<episode_id>/
  task.json
  1-1.jpg or 1-1.png
  1-2.jpg or 1-2.png
  1-2_label.jpg or 1-2_label.png  # present when an action label image is referenced
  ...

Each task.json contains:

{
  "phone": "device model",
  "os": "operating system",
  "screen_resolution": [width, height],
  "query": "high-level user instruction",
  "episode_id": "episode directory id",
  "data": [
    {
      "step": 1,
      "screenshot": "1-1.jpg",
      "observation": "textual screen observation",
      "plan": [
        {
          "description": "low-level action description",
          "type": "OPEN",
          "app": "app name"
        }
      ]
    }
  ]
}

The xml field and XML files are not included in this preview packaging.

Action Space

Action names have been normalized to the action space described for HyperTrack:

Normalized action Source local action Parameters
OPEN Open app, description
CLICK Tap position, box, times, description
SCROLL Swipe start_position, stop_position or end_position, box, description
TYPE Typing text, position, box, description
STOP End description

Coordinates and boxes are expressed in screenshot pixel coordinates. screen_resolution is stored as [width, height].

Filtering and Normalization

The preview subset was generated with:

python3 filter_hypertrack_episodes.py \
  --src test \
  --dst test_filtered \
  --report test_filter_report.json \
  --overwrite

The filtering method:

  • maps Open/Tap/Swipe/Typing/End to OPEN/CLICK/SCROLL/TYPE/STOP;
  • removes the per-step xml field and does not copy XML files into the preview output;
  • removes episodes whose referenced screenshot dimensions do not match screen_resolution;
  • removes episodes whose spatial action annotations fall outside the referenced screenshot pixel space;
  • removes episodes where SCROLL lacks a valid non-empty box;
  • keeps episodes with unreferenced extra label files, because they are not part of the referenced trajectory.

Removed episodes and exact reasons are recorded in data/test_filter_report.json.

Intended Uses

This preview subset is suitable for:

  • inspecting the HyperTrack-style static trajectory format;
  • building and debugging offline GUI-agent evaluators;
  • testing screenshot-plus-history prompt construction;
  • validating action parsing, coordinate handling, and action-grounding checks.

Out-of-Scope Uses

This preview subset should not be used as:

  • a replacement for the full HyperTrack dataset;
  • a complete benchmark leaderboard split;
  • evidence for full-dataset composition or performance claims;
  • a source for live app automation or account-affecting actions.

Privacy and Safety

Screenshots are privacy-filtered before release. Before external redistribution or publication, review screenshots, labels, text observations, and task instructions for privacy and compliance.

Mobile GUI-agent data can be misused to automate spam, fraud, unauthorized account activity, or attempts to bypass user safeguards. Use this subset for offline research, evaluation, and tooling validation.

Citation and Attribution

If using this preview subset, state that this is the filtered HF preview subset.

Downloads last month
80