duality-robotics/pose_estimation5.1

Pose Estimation Dataset

Duality AI released a 1000 image dataset for pose estimation, including images, masks, and labels -- and it's 100% free!

Just create an EDU account by clicking here.

This HuggingFace dataset is a 20 image and label sample divided into train and val folders, but you can get the rest at no cost by creating an EDU account. Once you verify your email, the link will redirect you to the dataset page.

What makes this dataset unique, useful, and capable of bridging the Sim2Real gap?

• The digital twins are not generated by AI, but instead crafted by 3D artists to be INDISTINGUISHABLE to the model from the physical-world objects. This allows the training from this data to transfer into real-world applicability
• The simulation software, called FalconEditor, can easily create thousands of images with varying lighting, posing, occlusions, backgrounds, camera positions, and more. This enables robust model training.
• The labels are created along with the data, and specifically the pose-estimation labels are easier to generate in simulation than in the physical world. This not only saves large amounts of time, but also ensures the labels are incredibly accurate and reliable.

This dataset is designed for pose estimation tasks, focusing on determining the position and orientation of an object in 3D space. The dataset includes images, masks, and labels for both training and validation, making it suitable for machine learning applications in 3D object tracking and computer vision. This dataset was generated using Duality.ai simulation software: FalconEditor. Try FalconEditor today to create data to be used for pose estimation on different objects.

Dataset Structure

The dataset has the following structure:

pose_estimation_dataset/
|-- train/
|   |-- images/
|   |   |-- 000000000.png
|   |   |-- 000000001.png
|   |   |-- ...
|   |-- labels/
|   |   |-- 000000000.txt
|   |   |-- 000000001.txt
|   |   |-- ...
|   |-- masks/
|       |-- 000000000.png
|       |-- 000000001.png
|       |-- ...
|-- val/
    |-- images/
    |   |-- 000000000.png
    |   |-- 000000001.png
    |   |-- ...
    |-- labels/
    |   |-- 000000000.txt
    |   |-- 000000001.txt
    |   |-- ...
    |-- masks/
        |-- 000000000.png
        |-- 000000001.png
        |-- ...

Components

1. Images: RGB images of the object in .jpg format.
2. Labels: Text files (.txt) containing 3D pose annotations. Each label file corresponds to an image and contains the following information:
   • Three position values [x, y, z] representing the object's location in 3D space.
   • Four quaternion values [qx, qy, qz, qw] representing the object's orientation in 3D space.
3. Masks: Binary masks (.jpg) highlighting the object’s silhouette in the image.

Usage

To use this dataset, load the images, labels, and masks for your pose estimation pipeline. Ensure that the corresponding image, label, and mask files share the same base filename.

Example

If you have train/images/image_1.png, the corresponding files will be:

• train/labels/image_1.txt
• train/masks/image_1.png

Label Format

Each .txt label file contains a single line in the following format:

x y z qx qy qz qw

Example:

0.12 0.45 0.78 0.0 0.707 0.0 0.707

Licensing

license: apache-2.0