Method

Monocular 3D Detection with Geometric Constraints Embedding and Semi-supervised Training [KM3D]
https://github.com/Banconxuan/RTM3D

Submitted on 22 May. 2020 16:12 by
peixuan li (University of Chinese Academy of Sciences)

Running time:0.03 s
Environment:1 core @ 2.5 Ghz (Python)

Method Description:
Abstract— In this work, we propose a novel one-stage and
keypoints-based framework for monocular 3D objects detection
using only RGB images, called KM3D-Net. 2D detection only
requires a deep neural network to predict 2D properties of
objects, as it is a semanticity-aware task. For image-based 3D
detection, we argue that the combination of the deep neural
network and geometric constraints are needed to estimate
appearance-related and spatial-related information synergisti-
cally. Here, we design a fully convolutional model to predict
object keypoints, dimension, and orientation and then combine
these estimations with perspective geometry constraints to com-
pute position attributes. Further, we reformulate the geometric
constraints as a differentiable version and embed it into the net-
work to reduce running time while maintaining the consistency
of model outputs in an end-to-end fashion. Benefiting from this
simple structure, we then propose an effective semi-supervised
training strat
Parameters:
DLA-34
Latex Bibtex:
@misc{2009.00764,
Author = {Peixuan Li},
Title = {Monocular 3D Detection with Geometric Constraints Embedding and Semi-supervised Training},
Year = {2020},
Eprint = {arXiv:2009.00764},
}

Detailed Results

Object detection and orientation estimation results. Results for object detection are given in terms of average precision (AP) and results for joint object detection and orientation estimation are provided in terms of average orientation similarity (AOS).


Benchmark Easy Moderate Hard
Car (Detection) 96.44 % 91.07 % 81.19 %
Car (Orientation) 96.34 % 90.70 % 80.72 %
Car (3D Detection) 16.73 % 11.45 % 9.92 %
Car (Bird's Eye View) 23.44 % 16.20 % 14.47 %
This table as LaTeX


2D object detection results.
This figure as: png eps pdf txt gnuplot



Orientation estimation results.
This figure as: png eps pdf txt gnuplot



3D object detection results.
This figure as: png eps pdf txt gnuplot



Bird's eye view results.
This figure as: png eps pdf txt gnuplot




eXTReMe Tracker