This repository is for using a 6DoF dataset which was labeled automatically using camera intrinsics and extrinsics.
This dataset has 9 different objects, in total there are 126.520 images - however only 59.376 are labeled due to the low quality of the remaining ones. These images were labeled automatically, this is why some of the cuboids may be a little bit off. The unlabeled images are in the Image archives, so there will be more images than ground truth files.
The camera parameters can be downloaded from: camera Parameters. The measurements of the objects can be downloaded from: measurements. The ground truths can be downloaded from: ground Truth.
The images can be downloaded from the following table:
| Class Name | Images |
|---|---|
| BeerMug | BeerMug |
| can | can |
| CircuitBoard | CircuitBoard |
| clock | clock |
| Coal | Coal |
| CoffeCup | CoffeCup |
| decorativeBulbOff | decorativeBulbOff |
| decorativeBulbOn | decorativeBulbOn |
| scotchTape | scotchTape |
The label files are .json files. The name of the files are the same as the images. The following keys are available in the json file:
| key name | Usage |
|---|---|
| frame_name | This is the name of the corresponding image |
| objectPoints3D | These are the 3D coordinates of the object in xyz order of the object |
| objectPoints2D | These are the transformed 2D coordinates on the image plane of the object |
| quaternion | This is the quaternion describing the rotation of the object. The first element is the imaginary part |
| translation | This is the translation vector of the object |
| camera_matrix | This is the intrinsic camera matrix, it will be the same over all images |
| discoef | These are the coefficients of the camera |
| paper3D | The objects are placed on a white paper, these are the 3D coordianates of the paper |
| paper2D | The objects are place on a white paper, these are the projected 2D coordinates of the paper |
If you find this dataset useful or use it in your research please cite:
@misc{khoefle2020,
author = {Kevin Hoefle},
title = {HomeBrewDataset6DoF},
year = {2020},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/kevinkit/HomeBrewDataset6DoF}},
}
If you want to support the hosting of this dataset and the development of further helping functions, you can donate