Yearning for Deep Learning
Yearning for Deep Learning

Competitive edge

Parameter / Capability

State of the art

TheWhollySee

Training set size (images)

20m

>1bn per year

Image labelling & classification

Manual

Automatic

Annotation time per image

42s

<0.1s / real time

Semantic segmentation

Infrastructure + objects Moving objects & agents

Acquisition cost
per 1000 images

$10

$1

Data modality

Mostly visible imagery

Visible, IR, night, LiDAR

Vehicle (re)cognition

Category

Category, motion, behaviour

Vehicle geometry & shape

Rectangular

Arbitrary - articulated trucks, trailers, tractors & outsized loads

Annotation dimensionality

2D

3D + orientation

Driving policy

Rule-based

Learned by example from optical flow

 
Print Print | Sitemap
© 2017 TheWhollySee Company Ltd. All rights reserved.