gmj03/DenseStep2M

This repository contains datasets for instructional video analysis tasks, focusing on dense temporal grounding and procedural semantics.

📢 Data Source & ID Mapping

All videos in this repository are sourced from the HowTo100M Dataset.

• The <video_id> used in all JSON files corresponds directly to the original YouTube Video ID.

---

1. DenseStep2M.json

Description

A large-scale dataset containing 1,885,504 detailed, temporally grounded instructional steps annotated across 99,248 high-quality instructional videos (totaling 7,212 hours). Constructed through a training-free automated pipeline leveraging multimodal foundation models (Qwen2.5-VL-72B and DeepSeek-R1-671B) to process noisy HowTo100M videos, achieving precise alignment between visual content, ASR transcripts, and procedural semantics.

Supplementary Files

• DenseStep2M_attr.json: Records the global statistics of the dataset.

  {
    "video_num": 99248,
    "step_num": 1885504,
    "word_per_step": 10.465,
    "step_per_video": 18.998
  }
  

• DenseStep2M_title.json: Contains high-quality titles generated for each video. These titles are produced by Qwen2.5-72B based on ASR transcripts generated by WhisperX.

  {
    "d0N6xOQCP6c": "How to Add a Ribbed Neckline to a Sweater",
    "BM0Bq7EBo4A": "How to Apply the Color Meshing Technique with Wool-E",
    ...
  }
  

Key Features

• Dense Temporal Annotations: 19.00 steps per video on average (vs. 10.6 in HowToStep), with 10.5 words per step describing operational purposes, durations, and quantities.
• Context-Aware Semantics: Captures abstract steps requiring cross-modal inference (e.g., "Prepare the wash bag" inferred from visual + ASR context).
• Temporal Precision: 46.25 R1-mIoU score in timestamp alignment, with 62.0% alignable and 56.1% well-aligned steps (vs. 21.3%/15.2% in original HowTo100M).
• Domain Diversity: Covers open-domain tasks including cooking, home gardening, personal care, and electronics assembly.

Structure (DenseStep2M.json)

{
  "<video_id>": [
    {
      "step": "string (instructional step description)",
      "start": "float (start time in seconds)",
      "end": "float (end time in seconds)"
    },
    ...
  ],
  ...
}

---

2. DenseCaption100.json

Description

A manually curated benchmark containing 100 high-quality instructional videos spanning 12 categories (cooking, home repair, crafts, etc.), specifically designed for evaluating dense video captioning and procedural step grounding. Each video is annotated with 11.0 detailed steps on average, incorporating both visual details (spatial information) and textual context derived from ASR transcripts.

Structure

{
  "<video_id>": [
    {
      "step": "string (instructional step description)",
      "start": "float (start time in seconds)",
      "end": "float (end time in seconds)"
    },
    ...
  ]
}