3D Pose Annotation for Fine-Grained Object Categories

Contents of Directories

• Anno3D: 3D pose annotations for StanfordCars/FGVC_Aircraft/CompCars
• CAD: Sample CAD models from ShapeNet for visualization use; list of CAD models used in our datasets
• Image: Sample images from the fine-grained datasets for visualization use
• Scripts: Sample scripts to visualize the annotations

Format of Annotations

The annotations are packed in pickle (.pkl) files, each of which include a dictionary of annotations,

dict[key] = anno,

where key is the filename, and anno is a dictionary containing the following fields:

• model_id: the corresponding fine-grained 3D CAD model ID from ShapeNet
• azimuth, elevation, theta: the rotation angles introduced in Section 3.2 of the paper
• distance: the parameter 'd' introduced in Section 3.2 of the paper
• u, v: the principal point (u, v) introduced in Equation (3)
• f: the focal length f in Equation (3)

Usage

To use the 3D pose annotations, you need to download the 2D images from the fine-grained recognition datasets as well as the 3D models from the ShapeNet dataset.

1. Download 2D Images

(1) StanfordCars:

http://imagenet.stanford.edu/internal/car196/cars_train.tgz

http://imagenet.stanford.edu/internal/car196/cars_test.tgz

After downloading, put the extracted directory cars_train and cars_test under Image/StanfordCars.

(2) FGVC_Aircraft:

http://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/archives/fgvc-aircraft-2013b.tar.gz

After downloading, put the extracted directory fgvc-aircraft-2013b/data/images under Image/FGVC_Aircraft.

(3) CompCars:

https://www.dropbox.com/sh/46de2cre37fvzu6/AABXtX8QqA6sx37k1IyZmNQ2a?dl=0

Please contact the data provider of CompCars ([email protected]) for the extraction password. After downloading, put the extracted directory data/image under Image/CompCars.

2. Download 3D CAD Models

(1) Cars (For StanfordCars/CompCars Use)

http://shapenet.cs.stanford.edu/shapenet/obj-zip/ShapeNetCore.v1/02958343.zip

After downloading, put the extracted 02958343 under CAD/ Please see CAD/StanfordCars3D.txt and CAD/CompCars3D.txt for the list of CAD models used in our 3D pose datasets.

(2) Aircraft (For FGVC_Aircraft Use)

http://shapenet.cs.stanford.edu/shapenet/obj-zip/ShapeNetCore.v1/02691156.zip

After downloading, put the extracted 02691156 under CAD/ Please see CAD/FGVC_Aircraft3D.txt for the list of CAD models used in our 3D pose datasets.

Visualization and Scripts

To visualize the samples, run cd Scripts && bash ./sample_visualize.sh

The usage of our annotations can be seen from the following python scripts,

• utils.py: containing the code for projection as well as converting the angles to rotation matrix and a simple .obj file reader
• sample_visualize.py: containing the code for visualization

Publications

• 3D Pose Estimation for Fine-Grained Object Categories
• Improving Annotation for 3D Pose Dataset of Fine-Grained Object Categories

Citation

Please cite the paper in your publications if it helps your research:

@inproceedings{wang20183d,
  title={3D Pose Estimation for Fine-Grained Object Categories},
  author={Wang, Yaming and Tan, Xiao and Yang, Yi and Liu, Xiao and Ding, Errui and Zhou, Feng and Davis, Larry S},
  booktitle={European Conference on Computer Vision Workshop},
  pages={619--632},
  year={2018},
  organization={Springer}
}

@article{wang2018improving,
  title={Improving Annotation for 3D Pose Dataset of Fine-Grained Object Categories},
  author={Wang, Yaming and Tan, Xiao and Yang, Yi and Li, Ziyu and Liu, Xiao and Zhou, Feng and Davis, Larry S},
  journal={arXiv preprint arXiv:1810.09263},
  year={2018}
}