• No results found

RoboSat.pink : Computer Vision framework for GeoSpatial

N/A
N/A
Protected

Academic year: 2021

Share "RoboSat.pink : Computer Vision framework for GeoSpatial"

Copied!
60
0
0

Loading.... (view fulltext now)

Full text

(1)

RoboSat.pink :

Computer Vision framework for GeoSpatial Imagery

@o_courtin

(2)
(3)

Earth Observation

Public Policy Monitoring

BioSphere Study

Human Development

Intelligence

Economic Intelligence

Emergency & Crisis

(4)
(5)

THE PERCEPTRON

A PROBABILISTIC MODEL FOR INFORMATION STORAGE AND ORGANIZATION IN THE BRAIN

Rosenblatt 1958

AI is an old Lady

(6)

Data V E C T O R Classification

World2Vec

(7)

Neurals Network Output Input Loss Function

Supervised Learning

(8)

Neurals Network Output Input Loss Function Trained Model Output

Supervised Learning

(9)

Polynom

Weighted Graph

Lossy Data Compression

Grey Box

(10)

« Everything is related to everything else,

but near things are more related than distant things. »

(11)

https://github.com/vdumoulin/conv_arithmetic

(12)

Unet Like Semantic Segmentation

(13)

Medical Imagery Autonomous Vehicle GeoSpatial Imagery

Computer Vision

(14)

Spatial is Special

MultiBands

Data Complexity

Huge Data

(15)

Spatial is Special

MultiBands

Data Complexity

Huge Data

Interoperability

FOSS4G

Open Data

(16)

RoboSat.pink

(17)

RoboSat.pink

Computer Vision ecosystem for GeoSpatial Imagery

DataSet Quality Analysis

Change Detection highlighter

Features extraction

(18)

RoboSat != RoboSat

.pink

https://github.com/mapbox/robosat/issues/184

(19)

RoboSat.pink Raster

Coverage WMS

(20)

RoboSat.pink GeoJSON PostGIS Raster Coverage WMS OSM PBF XYZ

(21)

RoboSat.pink GeoJSON PostGIS Raster Coverage WMS OSM PBF XYZ Masks Prediction

(22)

RoboSat.pink GeoJSON PostGIS Raster Coverage WMS OSM PBF XYZ Masks Prediction Masks Compare Vector Prediction Spotify differences areas

(23)

Compare Predicts against alternate datasets

Pink : Predicted by trained model

Green : Alternate dataset

(24)

Compare Predicts against alternate datasets

(25)

Compare Predicts against alternate datasets

Pink squares : Significant differences

(26)
(27)
(28)

Download WMS XYZ Rasterize GeoJSON Extract OSM pbf Cover XYZ Image Tile Rasters XYZ Label Subset Training DataSet Bbox XYZ dir

Data Preparation

Rasters PostGIS
(29)

Image Label Cross Entropy mIoU Lovasz

http://www.cs.toronto.edu/~wenjie/papers/iccv17/mattyus_etal_iccv17.pdf http://www.cs.umanitoba.ca/~ywang/papers/isvc16.pdf

https://arxiv.org/abs/1705.08790

(30)

https://arxiv.org/abs/1809.06839

https://github.com/albu/albumentations

(31)

https://arxiv.org/abs/1806.00844

(32)

Easy to deploy

(33)

https://github.com/datapink/robosat.pink/blob/master/docs/101.md

(34)

https://github.com/datapink/robosat.pink/blob/master/docs/101.md

101 RoboSat.pink Tutorial

In ~2-3h, on a GPU server:

- RoboSat.pink install

- Download data

- Data Preparation

- Training

- Inference

- Compare to OSM

- Vectorize result

(35)

More than an application, an easy to extent framework

(36)

Stacks

Proj 4 GEOS GDAL Rasterio CUDA cuDNN PyTorch NumPy OpenCV RoboSat.pink Pillow Shapely Osmium Mercantile SuperMercado Albumentations LeafLet + VectorGrid
(37)

So all you need is :

- Imagery

any file format readable by GDAL

- GPU

NVIDIA > 8Go RAM

- Initial skills

GeoSpatial Data and CLI fluency

- Labels

usualy the key point

(38)
(39)

Accurate Labels

(40)
(41)

From OpenData to OpenDataSet

(42)

Few performances Metrics

rsp train

~5 Mp/s, -per

epoch-rsp tile

~5 Mp/s

rsp predict

~5 Mp/s

rsp rasterize

~50 Mp/s

rsp vectorize

~50 Mp/s

8 cores CPU, single GPU (either RTX or V100), SSD 16 tiles = 4 Retina Tiles = 1Mp

(43)

How to scale it, or improve it again ?

rsp train

add more GPU,

reduce dataset redundancy,

improve model, loss or optimizer

rsp tile

add more CPU

use raster compression

rsp predict

export model to ONNX or JIT,

(44)

RTX ThreadRipper

Cost Effective GPU

WorkStation

Server

Cloud

(45)

Why performances matters ?

- Playful and Human Learning

- Time and money saver

(46)

Open Source

(47)

Request For Funding

- Increase prediction accuracy :

- on low resolution imagery

- even with few labels

- on network features

- instance segmentation

(48)

Request For Funding

- Increase prediction accuracy :

- on low resolution imagery

- even with few labels

- on network features

- instance segmentation

- Improve performances

- on Training

- on Inference

(49)

Request For Funding

- Increase prediction accuracy :

- on low resolution imagery

- even with few labels

- on network features

- instance segmentation

- Improve performances

- on Training

- on Inference

- Add support for :

- MultiClass classification

- Time Series Imagery

(50)

Why using Deep Learning for GeoSpatial ?

(51)

Why using Deep Learning for GeoSpatial ?

Easy to spotify at scale inconsistencies beetwen two datasets

If you provide accurate labels matching an imagery,

(52)

Why using Deep Learning for GeoSpatial ?

Easy to spotify at scale inconsistencies beetwen two datasets

If you provide accurate labels matching an imagery,

infere at scale on similar new imageries

(53)
(54)

Why using RoboSat.pink ?

GIS Standards compliancy

Ease Data Preparation

Build-in WebUI

Handle MultiBands Imagery and DataFusion

High Performances

Easy to deploy

Accurate (state of art Computer Vision)

Extensible by design

Open Source

Community

(55)
(56)

DIY Demo

www.datapink.com

(57)

Human Learning

http://www.math.ens.fr/~feydy/Teaching/culture_mathematique.pdf http://cs231n.stanford.edu/

https://neurovenge.antonomase.fr/NeuronsSpikeBack.pdf

(58)

Pattern

Extraction

Computer

Engineering

Data Expertise

Pattern

Extraction

Computer

Engineering

Data Expertise

(59)

Extract insights from GeoSpatial data with Deep Learning

@data_pink

(60)

RoboSat.pink

powered by

@data_pink

Take Away

- Industrial AI4EO Imagery framework available

- Performances already OK to use it for region or country

- No need anymore to be a Computer Vision expert to use it

- Plain OpenData can be use to train accurate model

https://github.com/mapbox/robosat https://github.com/datapink/robosat.pink https://github.com/datapink/robosat.pink/blob/master/docs/101.md https://github.com/datapink/robosat.pink/blob/master/docs/extensibility_by_design.md https://github.com/datapink/robosat.pink#geospatial-opendatasets https://github.com/datapink/robosat.pink/blob/master/docs/from_opendata_to_opendataset.md

References

Related documents