Home

Awesome

Render Synthetic Images with Part Segmentation Annotations Using CGPart

CGPart is a comprehensive part segmentation dataset that provides detailed annotations on 3D CAD models, synthetic images, and real test images. It involves 5 vehicle categories: car, motorbike, aeroplane, bus, and bicycle. Below are some example segmentation annotations from the dataset. You can find more information about CGPart from this page or from our paper.

<img src="[https://qliu24.github.io/udapart/images/cgpart_overview.jpg](https://github.com/qliu24/udapart/blob/gh-pages/images/cgpart_overview.jpg)" alt="cgpart_overview" width="750"/>

Requirements

Usage

Step1: Download the annotated 3D CAD models and setup the config files

Step2: Render the images (with keypoint annotations)

cd render_image_kp
python render_manual_anno_kp.py --obj_cls car --model_id 6710c87e34056a29aa69dfdc5532bb13

Step3: Render the depth maps and convert them into segmentation maps

cd render_seg
python render_manual_anno_parts.py --obj_cls car --model_id 6710c87e34056a29aa69dfdc5532bb13

Then run the render_seg/depth_to_semseg.py with proper arguments to generate the segmentation maps, for example:

python depth_to_semseg.py --obj_cls car --model_key sedan

Step4 (optional): Visualize the results

Example code is given in the visualization.ipynb notebook.

Citation

If you find this project helpful, please consider citing our paper.

@article{liu2019semantic,
  author    = {Liu, Qing and Kortylewski, Adam and Zhang, Zhishuai and Li, Zizhang and Guo, Mengqi and Liu, Qihao and Yuan, Xiaoding and Mu, Jiteng and Qiu, Weichao and Yuille, Alan},
  title     = {CGPart: A Part Segmentation Dataset Based on 3D Computer Graphics Models},
  journal   = {arXiv preprint arXiv:2103.14098},
  year      = {2021},
}