This is the repository for paper "ZoomNet: Part-Aware Adaptive Zooming Neural Network for 3D Object Detection".
The pixel-wise annotations on the KITTI trainval set is available via:
Sample code for processing the provided annotations.
importpickleimportnumpyasnpdefload_pickle(filename):
withopen(filename, 'rb') asf:
obj=pickle.load(f)
returnobjinfo=load_pickle(‘000002.pkl) # info.keys(): dict_keys(['data_idx', 'objects', 'calib', 'imw', 'imh', 'instL', 'instR’]). ‘InstL’ contains the pixel-wise inst_id(1-channel), depth(1-channel), part location (3-channel).# sample code for pixel-wise depth annotationpkl_objects=info['objects']
calib=info['calib']
f=calib['P'][0, 0]
bl= (calib['P'][0, 3] -calib['P3'][0, 3]) /ff_bl=f*blinst_map_left_=np.concatenate([np.expand_dims(el.toarray(), -1) forelininfo['instL']], axis=-1)
# convert depth to dispdispMapL=f_bl/inst_map_left_[:, :, 1].copy()
dispMapL[np.isinf(dispMapL)] =0print(dispMap.shape)The code for generating pixel-wise annotations and ZoomNet (pytorch) needs to be organised before release. A version on paddle-paddle is also expected to be released. However, I’m currently working on a workshop on CVPR and thus delayed the release of code. I'm sorry about that.
If you are benefited from this paper, please cite our paper as follows:
@inproceedings{xu2020zoomnet,
title={ZoomNet: Part-Aware Adaptive Zooming Neural Network for 3D Object Detection},
author={Xu, Zhenbo and Zhang, Wei and Ye, Xiaoqing and Tan, Xiao and Yang, Wei and Wen, Shilei and Ding, Errui and Meng, Ajin and Huang, Liusheng},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={2},
pages={7},
year={2020}
}