Method

MTrans-based Evidential Deep Learning autolabeler with Uncertainty Estimation capability [MEDL-U]
[Anonymous Submission]

Submitted on 9 Sep. 2023 12:21 by
[Anonymous Submission]

Running time:1 s
Environment:GPU @ >3.5 Ghz (Python)

Method Description:
We introduce an Evidential Deep Learning
(EDL) based framework that can be seamlessly integrated into
any learning-based 3D pseudo labeling system. Specifically,
we propose MEDL-U, an EDL framework based on MTrans,
which not only generates pseudo labels but also quantifies the
associated uncertainties. These uncertainties are then employed
to train probabilistic versions of existing 3D detectors.
Parameters:
We employed
a dropout rate of 0.4 and utilized the Adam optimizer
with a learning rate of 0.60e − 04. Autolabeler training is
conducted for 300 epochs, with a batch size of 5. Training
probabilistic 3D detectors using the generated pseudo labels
and uncertainties is conducted for 80 epochs.
Latex Bibtex:

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) 96.70 % 91.19 % 86.06 %
Car (Orientation) 94.27 % 86.11 % 79.84 %
Car (3D Detection) 85.43 % 75.56 % 68.79 %
Car (Bird's Eye View) 91.75 % 86.50 % 79.29 %
This table as LaTeX

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



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



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



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




eXTReMe Tracker