Skip to content

Latest commit

 

History

History
294 lines (233 loc) · 5.62 KB

slides.md

File metadata and controls

294 lines (233 loc) · 5.62 KB
theme addons transition author date image image-transform image-filter color-opacity
eavise
slidev-addon-animattr
fade
Tanguy Ophoff
26 June 2023
/earthobservation.jpg
scale(1.1)
grayscale(0.4)
0.5

Aerial Object Detection

PhD Defense

1052

satellites
observing
earth

<style> .quote { @apply text-8l; line-height: 1.3; text-transform: uppercase; font-weight: 300; text-align: center; @apply flex-(~ col align-center justify-center gap-1); position: absolute; inset: 0 50% 0 0; & p:first-child { @apply text-12l; } } </style>

layout: none

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1451187580459-43490279c0fa?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=1172&q=80)" imageFilter="grayscale(0.9)" v-click

How can we adapt detection algorithms
to work on remote sensing data?

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1506373823672-c9071ba1e864?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=1170&q=80)" imageFilter="grayscale(0.9)" v-click

How to combine color and depth data
to improve detection models?

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1517953377824-516f2dca803b?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=2089&q=80)" imageFilter="grayscale(0.9)" v-click

How much can we speed up our models
whilst maintaining the accuracy?

<style> .block p { @apply text-(center 20l); line-height: 1.2; } .slidev-vclick-target { transition: translate 200ms ease-in; } .block::after { transition: background-color 200ms; } .slidev-vclick-hidden { translate: 100% 0; } .block:has(+ .slidev-vclick-prior), .block:has(+ .slidev-vclick-current) { --color: black !important; color: rgb(150,150,150); } </style>

title: Section Teach layout: section number: 1

src: pages/01-teach.md color: var(--sns-desat-0) icon: teach

title: Section Satellite layout: section number: 2

src: pages/02-satellite.md color: var(--sns-desat-1) icon: satellite

title: Section Fusion layout: section number: 3

src: pages/03-fusion.md color: var(--sns-desat-2) icon: fusion

title: Section Plane layout: section number: 4

src: pages/04-plane.md color: var(--sns-desat-3) icon: plane

title: Section Speed layout: section number: 5

src: pages/05-speed.md color: var(--sns-desat-4) icon: speed

title: Section Conclusion layout: section

layout: none clicks: 7

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1451187580459-43490279c0fa?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=1172&q=80)" imageFilter="grayscale(0.9)" v-click="1"

How can we adapt detection algorithms
to work on remote sensing data?

sliding
window scene-specific
processing ResnetYolo

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1506373823672-c9071ba1e864?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=1170&q=80)" imageFilter="grayscale(0.9)" v-click="3"

How to combine color and depth data
to improve detection models?

transparent
fuse layer mid-late
fusion different
use cases remote
sensing

<Block class="w-full h-full flex-(~ items-center justify-center) p-5" :color-opacity="0.6" image="url(https://images.unsplash.com/photo-1517953377824-516f2dca803b?ixlib=rb-4.0.3&ixid=M3wxMjA3fDB8MHxwaG90by1wYWdlfHx8fGVufDB8fHx8fA%3D%3D&auto=format&fit=crop&w=2089&q=80)" imageFilter="grayscale(0.9)" v-click="5"

How much can we speed up our models
whilst maintaining the accuracy?

careful
selection constrainedness academic
4x industrial
15x
<style> .block p { @apply text-(center 20l); line-height: 1.2; } .block.slidev-vclick-target { transition: translate 200ms ease-in; } .block.slidev-vclick-hidden { translate: 100% 0; } .block::after { transition: background-color 200ms; } .block:has(+ .slidev-vclick-prior), .block:has(+ .slidev-vclick-current) { --color: black !important; color: rgb(200,200,200); } .answer { @apply absolute !p-5 flex-(~ justify-around items-center); inset: 0; span { @apply text-16l text-center; display: block; text-transform: lowercase; font-variant: small-caps; } } </style>