Method

BFF R-CNN: Balanced Feature Fusion for Object Detection. [BFF]


Submitted on 19 Nov. 2020 16:16 by
Yaze Li (BeiJing Union University)

Running time:8.4 s
Environment:4 cores @ 1.5 Ghz (Python)

Method Description:
Both End to Multi-Layers; BEtM,Multi-layer RoIE,
MRoIE;Recursive Feature Pyramid,RFP;
under faster rcnn resnet50, using mmdetection
framework
Parameters:
lr=0.2, epoch=5
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) 90.84 % 88.49 % 78.84 %
Pedestrian (Detection) 76.95 % 58.72 % 53.70 %
This table as LaTeX


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



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




eXTReMe Tracker