MVTec ITODD - A Dataset for 3D Object Recognition in Industry
Abstract
The MVTec Industrial 3D Object Detection Dataset (MVTec ITODD) is a public dataset for 3D object detection and pose estimation with a strong focus on industrial settings and applications.
The dataset consists of
- 28 objects and 3500 labeled scenes containing instances of these objects
- Five sensors (two 3D sensors and three grayscale cameras) observing each scene
More information can be found in this PDF file.
Download
Due to the size of the files, the download is split into multiple parts. All parts can be extracted into the same directory. The base package must be downloaded. Depending on which modalities your method operates on, the other packages can be downloaded as required.
Note that an evaluation on all data is preferred. For example, a method that uses 3D input data should be evaluated on both the high quality and the low quality 3D data, while a method that works on image data should be evaluated on all three cameras. However, you can also evaluate only on selected data, which will be mentioned in the result list and should be noted in any publication.
- Base Package (required): Contains the CAD models, calibration information, a few ground truth poses, and other information. This package always must be downloaded.
- 3D Data: Contains X, Y, and Z (i.e., range) images of each scene. Because the 3D data was acquired with stereo sensors, the download also contains the left camera image of each scene. The camera image has approximately the same viewpoint as the 3D / range data. Data of two sensors with different 3D quality are available:
- 2D Grayscale Images: Contains the images taken with the three grayscale cameras. Two images were taken with each: One with and one without a projected pattern that can be used, for example, for stereo reconstruction. Additionally, each image is provided in both its original form, which has optical distortions, and a rectified form.
- Download 2D Images without distortion, without projected pattern:
- Download 2D Images without distortion, with projected pattern:
- Download 2D Images with distortion, without projected pattern:
- Download 2D Images with distortion, with projected pattern:
Results
Here, you soon will find a list of all previous results.
Evaluate
While the base package contains a few ground truth poses, most poses are kept confidential to allow a fair comparison between different methods. To evaluate your results, please write them in the format described in the “result.txt” file of the base package, which also contains examples. The results essentially contain:
- Meta information about your method
- A list of rigid 3D transformations with all detections
To evaluate your results, pack all text files into a ZIP file and upload it in the form below. The results will be evaluated against the ground truth and sent back to you. Optionally, the results can be included in the result list.
Attribution
If you use the dataset in scientific work, please cite
Bertram Drost, Markus Ulrich, Paul Bergmann, Philipp Härtinger, and Carsten Steger. Introducing MVTec ITODD — A Dataset for 3D Object Recognition in Industry; in: IEEE International Conference on Computer Vision (ICCV), 2200-2208, October 2017.
License
The data is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0). For using the data in a way that falls under the commercial use clause of the license, please contact us.
Contact
If you have any questions or comments about the dataset, feel free to contact us via the form above as well.