ACF Based Region Proposal Extraction for YOLOv3 Network Towards High-Performance Cyclist Detection in High Resolution Images

A paper implementation or may be just aggregation of some open source implementations

Source: https://www.mdpi.com/1424-8220/19/12/2671/pdf

Why?

Idea, Code, Write-up

What?

Autonomous Driving Demo
Part of my slide on Topics in Autonomous Driving

YOLO network cannot achieve high precision when dealing with small size object detection in high resolution images.

Hypothesis

An effective region proposal extraction method for YOLO network to constitute an entire detection structure named ACF-PR-YOLO.

ACF-PR Region Proposal Generation Method

Generate large potential regions containing objects for the following deep network.

Source: https://www.mdpi.com/1424-8220/19/12/2671/pdf

Some extras about ACF

A bit about Merger and Extension of Bounding Boxes

One example of the process of merging bounding boxes. Source: https://www.mdpi.com/1424-8220/19/12/2671

YOLO Network for Cyclist Detection

Network Structure. Source: https://www.mdpi.com/1424-8220/19/12/2671/pdf

Post Processing

How?

It took me 3 months to implement Alpha Zero.

A slide of my proposed plan
Test on Inria Dataset
Train ACF
Modify Detector
ACF on Tsinghua-Daimler Dataset
Test on Tsinghua-Daimler Dataset
merge and extend bounding boxes
Results using merge and extend
Detection using ACF
Crop Patches
Save Patches after ACF execution
Complete Pipeline using ACF to Prepare Data for YOLO
cfg file used for training Tiny-YOLOv3
Test YOLO
Results after training Tiny-YOLOv3
Complete Process

Code

Demo

Cyclist Detection Demo

A great Failure

What’s Next?

References