FCOS - Fully Convolutional One Stage Detector (Aug 2019):
Object Detection without using anchor boxes or Region Proposal Network (RPN)
Timeline 👇
Paper: https://arxiv.org/pdf/1904.01355.pdf
This presentation was given at the IceVision Discord Forum where we organize weekly presentations featuring IceVision users and guests. Feel free to join if you are interested in learning about IceVision, the First Agnostic Computer Vision Framework
IceVision: https://airctic.com
To join: https://discord.gg/JDBeZYK
00:00 - Introduction
00:16 - Presentation Overview
02:50 - Introduction to Object Detection Architectures
08:07 - Architecture - Common Object Detector
10:35 - Feature Pyramid Network
12:37 - Faster R-CNN
13:18 - Anchor-based detectors drawbacks
16:45 - Anchor-free Detectors
18:27 - FCOS Motivations
20:00 - FCOS Architecture
23:21 - FCOS vs RetinaNet
26:26 - FCOS Approach
27:03 - FCOS Advantages
28:58 - Locations
32:12 - Center-ness
34:17 - Loss Functions
35:04 - Code Explore
39:14 - FCOS vs. Anchor-based Detectors
41:53 - FCOS vs. RetinaNet Performance
43:18 - Discussion
1:07:20 - End of Presentation
1:08:04 - Outro
🎧MUSIC:
Warriyo - Mortals (feat. Laura Brehm) [NCS Release]
Music provided by NoCopyrightSounds.
Watch: https://www.youtube.com/watch?v=yJg-Y5byMMw
Free Download / Stream: http://ncs.io/mortals
