The KITTI Vision Benchmark Suite

We introduce a simple framework for Monocular
DEtection with depth-aware TRansformer, named
MonoDETR. We enable the vanilla transformer to be
depth-aware and enforce the whole detection process
guided by depth. Specifically, we represent 3D
object candidates as a set of queries and produce
non-local depth embeddings of the input image by a
lightweight depth predictor and an attention-based
depth encoder. Then, we propose a depth-aware
decoder to conduct both inter-query and query-scene
depth feature communication. In this way, each
object estimates its 3D attributes adaptively from
the depth-informative regions on the image, not
limited by center-around features. With minimal
handcrafted designs, MonoDETR is an end-to-end
framework without additional data, anchors or NMS
and achieves competitive performance on KITTI
benchmark among state-of-the-art center-based
networks.

ResNEt50 + 3*Encoder + 3*Decoder

@article{zhang2022monodetr,
title={MonoDETR: Depth-aware Transformer for
Monocular 3D Object Detection},
author={Zhang, Renrui and Qiu, Han and Wang, Tai
and Xu, Xuanzhuo and Guo, Ziyu and Qiao, Yu and Gao,
Peng and Li, Hongsheng},
journal={arXiv preprint arXiv:2203.13310},
year={2022}
}