VerboseTechLabs/textile-manufacturing-egocentric-sample

🧵 Textile Manufacturing — Egocentric Video Dataset (Sample)

This dataset is part of a larger collection of egocentric activity datasets by Verbose Tech Labs LLP. If you want the full dataset, or want access to more categories? Get in touch with us:

• 📞 Phone: +91 7672 000 500
• 💬 WhatsApp: +91 7672 000 500
• 📧 Email: Hello@VerboseTechLabs.com
• 🌐 Website: VerboseTechLabs.com
• 🔗 More datasets: kaggle.com/verbosetechlabsllp

---

Dataset Summary

First-person point-of-view (POV) video recordings from textile manufacturing operations, captured on real factory floors. This is a sample release showcasing the format and quality of our larger textile industry dataset collection.

Dataset Statistics

Metric	Value
Total clips	11
Total duration	~30.5 minutes
Total size	~1.04 GB
Activity class	textile_manufacturing
View type	Egocentric (first-person)
Video format	MP4
Frame rate	30 fps
Resolution	1080p

Supported Tasks

• Video classification — classify textile manufacturing activities
• Action recognition — recognize textile industry actions
• Fine-grained textile activity classification
• Worker productivity and time-motion analysis
• Machine operation understanding (looms, knitting machines, dyeing units)
• Ergonomics research in textile industry
• Assistive robotics for textile factories
• Quality control AI training
• Industrial AI for textile automation

Dataset Structure

Folder Structure

textile-manufacturing-egocentric-sample/
├── videos/
│   ├── textile_01.mp4
│   ├── textile_02.mp4
│   ├── textile_03.mp4
│   ├── textile_04.mp4
│   ├── textile_05.mp4
│   ├── textile_06.mp4
│   ├── textile_07.mp4
│   ├── textile_08.mp4
│   ├── textile_09.mp4
│   ├── textile_10.mp4
│   └── textile_11.mp4
├── metadata.csv
└── README.md

Data Fields

The metadata.csv file contains the following columns:

Column	Type	Description
file_name	string	Relative path to the video file
clip_id	string	Unique identifier (e.g., TXT_001)
activity	string	Main class: textile_manufacturing
sub_activity	string	Fine-grained label
duration	string	Human-readable duration (HH:MM:SS)
duration_seconds	integer	Duration in seconds
file_size_mb	float	File size in megabytes
recording_date	date	Recording date (YYYY-MM-DD)
resolution	string	Video resolution
fps	integer	Frames per second
view_type	string	Camera view type (egocentric)
notes	string	Additional context

Clip Overview

Clip ID	File	Duration	Size
TXT_001	textile_01.mp4	00:00:30	24 MB
TXT_002	textile_02.mp4	00:03:00	104 MB
TXT_003	textile_03.mp4	00:03:00	104 MB
TXT_004	textile_04.mp4	00:03:00	104 MB
TXT_005	textile_05.mp4	00:03:00	104 MB
TXT_006	textile_06.mp4	00:03:00	104 MB
TXT_007	textile_07.mp4	00:03:00	104 MB
TXT_008	textile_08.mp4	00:03:00	104 MB
TXT_009	textile_09.mp4	00:03:00	104 MB
TXT_010	textile_10.mp4	00:03:00	104 MB
TXT_011	textile_11.mp4	00:03:00	104 MB

Activity Coverage

The dataset captures diverse textile manufacturing activities from real mill floors, spanning operations across the textile production pipeline — spinning, weaving, knitting, dyeing, printing, finishing, and quality control.

Usage

Load with 🤗 datasets library

from datasets import load_dataset

dataset = load_dataset("verbosetechlabsllp/textile-manufacturing-egocentric-sample")
print(dataset)

Load metadata directly with Pandas

import pandas as pd

df = pd.read_csv("hf://datasets/verbosetechlabsllp/textile-manufacturing-egocentric-sample/metadata.csv")
print(df.head())
print(f"Total duration: {df['duration_seconds'].sum() / 60:.1f} minutes")

Download a specific video

from huggingface_hub import hf_hub_download

video_path = hf_hub_download(
    repo_id="verbosetechlabsllp/textile-manufacturing-egocentric-sample",
    filename="videos/textile_02.mp4",
    repo_type="dataset"
)
print(f"Video downloaded to: {video_path}")

Extract sample frames

import cv2, os

def extract_frames(video_path, out_dir, every_n_seconds=5):
    os.makedirs(out_dir, exist_ok=True)
    cap = cv2.VideoCapture(video_path)
    fps = cap.get(cv2.CAP_PROP_FPS)
    frame_interval = int(fps * every_n_seconds)
    count, saved = 0, 0
    while True:
        ret, frame = cap.read()
        if not ret: break
        if count % frame_interval == 0:
            cv2.imwrite(f"{out_dir}/frame_{saved:04d}.jpg", frame)
            saved += 1
        count += 1
    cap.release()
    return saved

Data Collection

• Camera view: First-person / egocentric (head-mounted or chest-mounted)
• Environment: Real textile mill / factory floor
• Lighting: Industrial factory lighting
• Audio: Included in MP4 (ambient loom, machine, and worker sounds — usable for multimodal research)
• Recording period: February 2025 – May 2026

Licensing Information

CC BY 4.0 — Free for research and commercial use with attribution.

Citation

@dataset{textile_manufacturing_egocentric_2026,
  title  = {Textile Manufacturing — Egocentric Video Dataset (Sample)},
  author = {Verbose Tech Labs LLP},
  year   = {2026},
  url    = {https://huggingface.co/datasets/verbosetechlabsllp/textile-manufacturing-egocentric-sample}
}

More Datasets from Verbose Tech Labs

This dataset is part of a larger collection of egocentric activity datasets covering:

• 👕 Clothing industry manufacturing
• 🍳 Cooking & food preparation
• 🧹 Household cleaning tasks
• 🏭 Manufacturing unit workflows (sample)
• 🛠️ Skilled commercial work (sample)
• 🧵 Textile manufacturing (this — sample)
• ...and more categories in development

🔗 Browse all our datasets: kaggle.com/verbosetechlabsllp | huggingface.co/verbosetechlabsllp