Datasetpublic

Ji0134ch/Synthetic-Glass-Transparent-Packaging-Dataset-Sample

Transparent Packaging & Glass Benchmark Watch our benchmark breakdown: Why Object Detection Fails on Glass. Synthetic Transparent Glass & Packaging Dataset A photorealistic synthetic computer vision dataset for transparent glass and packaging object detection and instance segmentation. The dataset is designed for models dealing with challenging transparent and reflective materials, including transparency, reflections, refractions, specular highlights, and harsh… See the full description on the dataset page: https://huggingface.co/datasets/Ji0134ch/Synthetic-Glass-Transparent-Packaging-Dataset-Sample.

sourceHugging Faceotherupdated 24d agoView on Hugging Face

0likes86downloads

Dataset Card

banner1

Transparent Packaging & Glass Benchmark

![Watch the Benchmark Overview](https://www.youtube.com/watch?v=aKlNZl4Qsro)

Watch our benchmark breakdown: Why Object Detection Fails on Glass.

Synthetic Transparent Glass & Packaging Dataset

A photorealistic synthetic computer vision dataset for transparent glass and packaging object detection and instance segmentation. The dataset is designed for models dealing with challenging transparent and reflective materials, including transparency, reflections, refractions, specular highlights, and harsh lighting conditions.

Dataset at a Glance

Property	Details
Dataset type	Synthetic computer vision dataset
Objects	Transparent glass and packaging
Images	240
Resolution	1024 × 1024
Training split	200 images
Validation split	40 images
Detection annotations	YOLO bounding boxes
Segmentation annotations	YOLO instance segmentation polygons
Image format	JPEG
Annotation format	YOLO `.txt`
Ground truth	Automatically generated from 3D scene data
Primary tasks	Object detection and instance segmentation
Material challenges	Transparency, reflections, refractions, specular highlights

This repository contains a 5-image sample dataset with dual ground-truth annotations so you can test pipeline compatibility instantly.

🛒 Download the Full Production Dataset (240 Images)

To fine-tune or evaluate your models on the complete dataset, access the full pack on Gumroad:

👉 [Download Full Dataset on Gumroad ($49)](https://5940734191807.gumroad.com/l/vjlhry)

Full Dataset Specifications

—240 Renders @ 1024x1024: 200 Train / 40 Val native resolution split.
—Dual Ground-Truth Annotations: Includes both 2D Bounding Boxes (labels_box/) and Multi-Point Instance Polygons (labels_segment/).
—Automatically Generated Ground Truth: Bounding boxes and segmentation polygons are generated directly from the underlying 3D scene, eliminating manual annotation work.
—Commercial Single-Team License: Royalty-free commercial usage rights to train, evaluate, and deploy models.

—indoor-pack : https://huggingface.co/datasets/Ji0134ch/synthetic-transparent-objects-indoor

Repository Structure

text

.
├── data/
│   ├── images/          # Sample 1024x1024 JPEG renders
│   ├── labels_box/      # YOLO format bounding box annotations (.txt)
│   └── labels_segment/  # YOLO format instance segmentation polygons (.txt)
├── src/
│   └──visualize_sample.py  # Script to draw bounding box and polygon overlays
└── README.md

Applications

This dataset is designed for computer vision tasks involving transparent and reflective objects, including:

—Transparent glass object detection
—Transparent packaging detection
—YOLO object detection
—Instance segmentation
—Synthetic-to-real computer vision experiments
—Robotics and 3D vision research