we use
original data
to create value

Our mission is to produce original and actionable data delivered in tailor made platforms.

learn more

in a data-driven world

We are a company committed to a major revolution : using data as a raw material to bring about radical changes.

in a data opening global movement

unique opportunities and new challenges arise every day

our mission is to produce original data

producing unique data is our way of enriching this huge pool of opportunities

by developing artificial intelligence for good

creating original data requires permanent investment in artificial intelligence

and building efficient and robust digital infrastructures

our objective is to build infrastructures from original data

In the energy, renovation, construction, insurance, green finance and culture fields, we gather technical and scientific partners in public and private projects to build digital platforms that enable environmentally friendly economic development.

our expertise

we produce original data

We collect available, non-personal data to generate a unique data library of original data using our AI tools.

Open data is an unprecedented opportunity for value creation. Building unique datasets and making them actionable allow for the emerging of new opportunities.

producing original data

using our AI tools and from images, texts and structured data, we create original data sets

connecting heterogeneous datasets

through our relationship engine and the creation of datasets, the data we produce becomes relevant, categorised by business sectors, with geo-tracking at the right level of granularity

making data actionable

Through integration in information systems of our clients or through development of tailored platforms, data is a tool for decision making or for building new opportunities.

Our datasets are actionable according to the constraints and needs of our clients and partners.

business opportunity sheets and clusters

we deliver opportunity sheets and cluster sheets directly in the tools of our clients through tailored solutions and interfaces.

tailored platforms

we develop platforms allowing our partners to use by themselves our databases and AI tools to generate their own sheets and clusters to create value, in their current projects or in new ones.

  • Using our AI tools and from images, texts and structured data, we create original data sets.

  • Through our relationship engine and the creation of datasets, the data we produce becomes relevant, categorised by business sectors, with geo-tracking at the right level of granularity

  • We deliver opportunity sheets and cluster sheets directly in the tools of our clients through tailored solutions and interfaces.

  • We develop platforms allowing our partners to use by themselves our databases and AI tools to generate their own sheets and clusters to create value, in their current projects or in new ones.


Our specific platforms allow us to gather complementary key actors around relevant data, in order to generate economies of scale and create new and better services. The platform hence becomes a genuine asset, a mutualized infrastructure dedicated to the exploitation and the creation of value upon the relevant data.

We accelerate exchange, reuse and valorisation of data through tailored platforms.



Grégory Labrousse

Pierre Lescure
Board Member

Lila Tretikov
Board Member

Emmanuel Bacry
Scientific Committee President

Pierre Alain De Malleray
Board Member

Eric Euvrard
Board Member

Raoul Saada
Finance Operations

Sebastián Sachetti
Business Development

Nicolas Berthelot
Lead Data Strategy

Louis Petros
Lead Knowledge Strategy

Gaël Grasset
Lead Product

Servane Khaouli
HR Business Partner

Alexandre Bacchus
Data Scientist

Duccio Piovani
Data Scientist

Charles Hutin-Persillon
Data Strategist

Aymeric Flegeo
Data Engineer

Guillaume Larcher
Developer Product

Vincent De Chillaz
Data Analyst - Quality

Alexander Usoltsev
Computer Vision Scientist

  • Grégory Labrousse

    CEO & Founder

    President of nam.R, serial entrepreneur, specialised in environment and cost reduction consulting. Founder of a group specialised in green finance, agro-ecology and eco-tourism.

  • Pierre Lescure

    Co-Founder & Board Member

    Author of the 2013 Digital Economy report, which underlined, among other things, the importance of open data, Pierre was involved in the creation of many companies, keeping in mind to always be at the forefront of the technologies involved (Molotov.tv, Canal+). Pierre is also a board member of the Kudelski cybersecurity group and the Lagardère group, and president of the Cannes Film Festival.

  • Lila Tretikov

    Co-Founder & Board Member

    Lila is Chief Executive Officer of the Terrawatt initiative, the “armed wing” of the International Solar Alliance, launched with the support of the President of the French Republic. Former President of the Wikimedia Foundation, where she initiated Wikipedia’s ambitious missions in digitalisation and artificial intelligence, Lila is a recognized expert in machine learning, with an impressive career (Chief Product Officer of SugarCRM, CEO of Raskspace...). Her name featured in Forbes list of The World's 100 Most Powerful Women (2014) and in San Francisco Chronicle’s “21 Most Powerful Women in Bay Area Technology”. She was awarded a Stevie Award for Woman in Business, and is a member of the Young Leaders of the World Economic Forum.

  • Emmanuel Bacry

    Co-Founder & Board Member

    CNRS Research Director at Paris Dauphine University. Professor and head of the “Data Science & Big Data” Initiative at Ecole Polytechnique. In charge of the "Big Data" processing of the French Social Security database.

  • Pierre-Alain De Malleray

    Board Member

    A former student of the ENA (promotion Senghor) and Ecole Polytechnique, an Inspector of finance, Pierre-Alain was successively ministerial advisor and managing director of MutRé. A recognized data specialist and president of insurance broker Santiane, he joined nam.R’s board in 2017.

  • Eric Euvrard

    Board Member

    Director and President of the Audit Committee of the Atari group, he began his career at Arthur Andersen where he participated in the development of the “Restructuring” practice. He then joined Lucien Deveaux for the takeover of the Bidermann Group where he directed the turnaround, before launching an Internet start-up, which he sold in 2002. It was then that he took over Gigastore, a non-food discount branch, through an LBO, that he directed until its sale in 2008. Erick manages a consulting firm specialising in mutation phases and co-leads a training group.

  • Raoul Saada

    Finance Operations

    A graduate from Polytechnique, an expert in financial and restructuring intermediation, Raoul participated in the creation of the equity broker Finacor and the implementation of the EuroMTN market. A specialist of green finance, he has been actively involved in many structuring transactions in the sector.

  • Sebastián Sachetti

    Business Development

    A graduate of the Ecole Nationale d’Administration, Sebastian is an outstanding polyglot (speaking French, English, Italian, Spanish and Portuguese) who has successively held leading positions in the private and the public sectors, combining the worlds of finance, culture et IT. Responsible for the conception and the implementation of the Pass Culture before joining our team, Sebastian is currently developing the new nam.R platforms.

  • Nicolas Berthelot

    Lead Data Strategy

    A recognized expert and passionate person, involved in the processing of socio-economic databases since his early childhood, Nicolas joined nam.R from the beginning to build his Data Library, a project that is unique in its size and innovation. A graduate of Sciences Po in Paris, Nicolas is one of the first in France to graduate in Data Strategy (Sorbonne University - UPMC). At nam.R, he leads the "Data Sourcing" and "Data Strategy" teams.

  • Louis Petros

    Lead Knowledge Strategy

    Passionate about economics, strategy and public affairs, Louis joined nam.R at the launch of the project, after working at EDF Trading in London, the French National Assembly and the French Ministry of Defense. A graduate in Political Science at the IEP in Strasbourg and holder of a Master's degree from the London School of Economics, Louis is responsible for the teams in nam.R's "Solutions" department, where he identifies and builds the solutions in which nam.R is involved.

  • Gaël Grasset

    Lead Product

    A Data Scientist and with an education in sociology, Gaël joined nam.R at the beginning of the project after graduating in statistical engineering at ENSAE and holding a Master's degree in sociology at Sciences Po Paris. Data Scientist, manager, product manager... Gaël has always been at the heart of nam.R's evolution, thanks to his technical expertise in Artificial Intelligence, which he developed at Oscaro. He is in charge of nam.R's “Product” teams.

  • Servane Khaouli

    HR Business Partner

    A graduate in history and project management, with many experiences in culture, communication and SME management, Servane has been involved with nam.R since its creation. At the heart of her responsibilities, combining human resources and office management, Servane supports the development of the various departments.

  • Alexandre Bacchus

    Data Scientist

    Holder of a doctorate in electrical engineering and a recognized expert in artificial intelligence, particularly in energy, Alexandre has worked successively on innovation and data projects at EDF and Enedis. Trained in agile project management, he joined nam.R's data science teams to help build the Digital Twin and organise team management.

  • Duccio Piovani

    Data Scientist

    After obtaining his PhD in complex systems at the prestigious Complexity Center of Imperial College London, Duccio has worked successively as a data scientist and researcher at the Center for Advance Spatial Analysis at Imperial College. Duccio joined nam.R to design the proprietary algorithms for nam.R.

  • Charles Hutin-Persillon

    Data Strategist

    A graduate of the IEP of Grenoble, Charles is pursuing a research career in Social Science, Political Science and International Relations. Passionate about environmental issues and the use of statistics, he has gradually turned to Data to become, in September 2017, Data Strategist at nam.R.

  • Aymeric Flegeo

    Data Engineer

    Aymeric graduated in Data Science at Télécom Paris, trained in Data Engineering at Renault, before joining nam.R. He joined the technical team in 2017 as a Data Engineer.

  • Guillaume Larcher

    Developer Product

    A graduate of the prestigious Master MVA program at ENS Paris-Saclay and Ecole Centrale of Lille, Guillaume joined nam.R in April 2017 for his final year internship. A jack of all trades, specialised in computer vision and development, he is working - after several months spent setting up image recognition algorithms - with the Product team to develop nam.R products.

  • Vincent De Chillaz

    Data Analyst - Quality

    An engineer from the Ecole Centrale of Lyon and a specialist in energy and climate, Vincent spent several years working for the renowned Carbone 4 firm, after working at Schlumberger, Vinci and Citeo. Lecturer at ESTP and passionate about data projects, Vincent joined nam.R as Product owner for the Digital Twin.

  • Alexander Usoltsev

    Computer Vision Scientist

    Alexander holds a Master's degree in Biometrics and first worked as a Research Engineer before joining nam.R in early 2018. As part of the Computer Vision team, Alexander is working on satellite images that he segments to identify different objects of interest to the start-up, particularly regarding solar energy.

  • And also...

    Discover the other nam.R team members

Clément Perny

Data Scientist

Clément is a final year Data Science Master student at Grenoble ENSIMAG engineering school and is currently working at nam.R as part of a NLP internship at the end of his studies. He works with the Data Science team on several projects.

Dina Khattab

Data Scientist

A Master 2 student in Data Science at Sorbonne University, Dina joined nam.R as part of a final year internship as a Data scientist. Dina has found in nam.R’s project and its use of open data in the context of the ecological transition a real opportunity to work on useful and concrete projects with the help of data science tools and technologies.

Hermès Martinez

Data Scientist

A graduate in language science from Paris Diderot University, Hermès is passionate about recent advances in Natural Language Processing. He puts his passion and expertise at the service of the nam.R project by manipulating unstructured textual data.

François Andrieux

Data Scientist

A graduate engineer from the ESIEA (École supérieure d'informatique, électronique, automatique), François is a true machine learning enthusiast, a passion he nurtures in a blog praised by amateurs, in parallel with his role as OpenClassRoom tutor. He joined nam.R in 2018 as a Data Scientist.

Paul-Louis Barbier

Full Stack Developper

A graduate in computer science and information systems, Paul-Louis was a Backend developer and Lead Dev in a startup before joining nam.R in May 2018. He is in charge of developing the Data Library for the Data Strategy team.

Sébastien Ohleyer

Computer Vision Scientist

A graduate of the Ecole Centrale of Lille in Data Analysis and holder of a Research Master in Mathematics from the University of Lille 1, Sébastien then joined the Master MVA of the Ecole Normale Supérieure Paris-Saclay. It is within the framework of this Master's degree that he joined nam.R as a Computer Vision Scientist.

Bastien Hell

Computer Vision Scientist

After graduating from an engineering school, Bastien joined the Institut national de l'information géographique et forestière (IGN) to carry out image processing and deep learning applied to geographical information. In early 2018, he joined nam.R as a Computer Vision Scientist.

Florentin Fromont

HR Administrator

Florentin holds a Master's degree in HR Management and Sustainable Performance and joined nam.R in February 2018. He is the HR assistant. His role is to ensure the administrative management of HR, set up a follow-up of personnel files, and intervene upstream in the recruitment process.

Frédéric Maison

Office Manager

Frédéric is a development manager for Geo PLC. Frédéric uses his experience in office management to support nam.R's growth and the sustainability of its structure.

Jules Robial

Art Director

A graduate of the Métiers d'Art at the Estienne school of art in Paris, Jules has a graphic designer and a typographer background. Working at nam.R since the end of 2017, Jules has been in charge of the startup's entire visual identity, on all its graphic or communication media, in order to develop the startup's image.

Adèle Bayart

Community Manager

After four years of studying communication and strategy at EFAP, Adèle had her first experience in community management in the luxury sector. She joined the nam.R project at the end of 2017 as Community Manager. As such, she’s responsible for communication on social networks and developing the visibility of the startup online.

Valentine Lambolez

Data Engineer

Valentine holds a Master's degree in Statistics and Socio-Economic Informatics from Université Lumière - Lyon II, where she specialised in Data Engineering. She joined Deepki, where she was in charge of the development and optimal implementation of customer-specific statistical models. She joined nam.R in 2018 where she is working as a Data Engineer.

Frédéric Kingue Makongue

Data Strategist

A graduate of the CNAM and the University of Bordeaux, specialised in digital archiving, Frédéric strengthens the Data Strategy team thanks to his expertise in ontology construction, document set structuring and massive data corpus analysis.

Corentin Louison

Business Analyst

Currently in his first Master year of Science, International strategy & influence at SKEMA BS, Corentin joined nam.R in July 2018 for a 6-month gap year internship in the Solution division and is participating in technological survey and ecosystem missions.

Juliette Cocault

Business Analyst

After obtaining a Bachelor in Commerce from McGill University in Canada, Juliette is pursuing her studies at EDHEC. As part of her Master's degree, Juliette joined nam.R for a 6-month internship in Strategy. Within nam.R, she is working on the company’s strategic monitoring, the implementation of an automatic monitoring tool, and the development of performance indicators for energy retrofitting.

our partners

For the development of our artificial intelligence tools and the definition of business rules applicable to various sectors, we rely on private and public structures such as research laboratories and business experts.


Advanced Aerial Imagery Analysis with Deep Neural Networks Explained in 5 Minutes.

There is no secret that when dealing with aerial images the best state-of-the-art results are achieved with deep learning models which come at the cost of their complexity. At the same time thanks to the Open Data, we can explore in a creative way even the most sophisticated techniques.

Read More

Developing a Complex Computer Vision System, a Case Study : Solar Panels Equipped Roofs.

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Duis vitae eros vel felis euismod malesuada a vitae est. Praesent fringilla nisi dolor, et venenatis justo pellentesque ut. Fusce ultrices eleifend augue. Curabitur sollicitudin elit vitae iaculis gravida.

Read More


Vivatech 2018

Paris, France - 05/2018


Emmanuel Macron était à #vivatech2018. Grégory Labrousse a pu lui présenter nam.R qui incarne parfaitement les valeurs d'une startup d'intérêt général.

Vivatech 2018

Paris, France - 05/2018


Un stand aux couleurs de nam.R pour servir de scène aux pitch des 3 jours de #vivatech2018. #DIY

AI for Good 2018 - ITU

Geneva, Switzerland - 05/2018


Conférence #AIforgood2018

Data Science Summer School

Paris, France - 07/2018


Duccio et Sacha sont toute la semaine de la Data Science Summer School pour présenter les projets de nam.R. #DS3

GeoDataDays 2018

Le Havre, France - 07/2018


@namr_france était présent aux #GeoDataDays2018 organisés par @afigeo_asso_fr et @DecryptaGeo. Grand rendez-vous des producteurs et utilisateurs des données géographiques. #opendata #geodata

Web Summit 2017

Lisboa, Portugal


Président Hollande talks with Gregory Labrousse about the Open Data révolution. François Hollande was one of Open Data’s pioneers alongside President Barack Obama.

Web Summit 2017

Lisboa, Portugal


Meeting with Professor Mohan Munasinghe

, Nobel Prize for Peace dedicated to his contribution to define the Sustainable Development Goals. nam.R is a key actor in the SDG working groups.

Data Science Summer School

Paris, France - 07/2018


Grégory rencontre Yann LeCun à la Data Science Summer School. L'occasion de présenter le travail de nam.R pour mettre la #datascience au service de la transition écologique. #iaforgood

Hackathon DataEnergie 2017

Paris, France - 06/2017


nam.R est lauréat du Hackathon DataEnergie 2017 organisé par @rte_france @GRTgaz @enedis @GRDF @Etalab et @LIBERTE_LL. #hackathon #victory

Websummit 2017

Lisbon, Portugal - 11/2017


L'équipe nam.R au complet (ou presque !) s'était rassemblée à Lisbonne à l'occasion du #websummit2017.

Websummit 2017

Lisbon, Portugal - 11/2017


Le stand de nam.R au #websummit2017, un lieu pour découvrir le projet ambitieux de nam.R et son #digitaltwin.

Websummit 2017

Lisbon, Portugal - 11/2017


A la rencontre d'@enigma_data, société majeure de l'#opendata aux Etats-Unis. Une curiosité toute particulière pour l'immense dépôt de données qu'ils ont constitué : Enigma Public. #opendata

Websummit 2017

Lisbon, Portugal - 11/2017


Le stand en construction de nam.R, le projet prend vie au #websummit2017.

Dreamforce 2017

San Francisco, U.S.A - 11/2017


nam.R était à #dreamforce2017, la grand messe de Salesforce. Au programme, toutes les déclinaisons du préfixe my : #myEinstein, #mySalesforce, #myTrailhead !

World Efficiency Solutions 2017

Paris, France - 12/2017


@GrassetGael présente nam.R au #WorldEfficiencySolutions2017.

World Efficiency Solutions 2017

Paris, France - 12/2017


Le stand de nam.R au #WorldEfficiencySolutions2017, événement complémentaire au #OnePlanetSummit !

World Efficiency Solutions 2017

Paris, France - 12/2017


@LouisPetros est présent sur le stand pour présenter le projet de nam.R pour la #transitionécologique. #WorldEfficiencySolutions2017

World Efficiency Solutions 2017

Paris, France - 12/2017


@g_labrousse et @LouisPetros présentent à la ministre @brunepoirson le projet de nam.R pour la #transitionécologique. #WorldEfficiencySolutions2017

Big Data Paris 2018

Paris, France - 03/2018


@GrassetGael représente @namr_france pour les finales du Trophées Big Data à @BigDataParis

Big Data Paris 2018

Paris, France - 03/2018


@Nicolas_data présente la #DataLibrary et son usage à #BigDataParis2018. #opendata #satellite #aerialimagerey

Big Data Paris 2018

Paris, France - 03/2018


@g_labrousse présente le projet nam.R à une audience nombreuse rassemblée à #BigDataParis2018

Big Data Paris 2018

Paris, France - 03/2018


Lila Tretikov, CEO de Terrawatt Initiative & co-fondatrice de nam.R à Big Data Paris

Observatoire de l'Open Data

Paris, France - 04/2018


nam.R est fier d'avoir participé à la création de l'Observatoire de l'Open Data en partenariat avec @OpenDataFrance, @Etalab, @caissedesdepots, @sciencespo. #opendata

Vivatech 2018

Paris, France - 05/2018


La Team nam.R est présente en force à #vivatech2018. Les curieux de #DigitalTwin, d'#ODD et d' #AIforgood sont venus nombreux pour discuter des projets de @namr_france.

AI for Good 2018 - ITU

Geneva, Switzerland - 05/2018


Gaël à la conférence de l'International Telecommunications Union pour l'#IAforGood

Meet-up Green Tech Verte

Paris, France - 05/2018


@g_labrousse présente nam.R au Meet-up de la GreenTechVerte

Conférence Comité21

Paris, France - 06/2018


Conférence #anthropocene & #ia du @Comite21, avec @Bettina_Laville & @LouisPetros de @namr_france

Toulouse Space Show 2018

Toulouse, France - 06/2018


@Charles_data et @GuillaumeLarch sont présents au #ToulouseSpaceShow2018. Ils peuvent y échanger sur notre usage de la donnée satellite aux experts du secteurs, producteurs comme utilisateurs. #satellite #newspace #opendata

Data Science Summer School

Paris, France - 07/2018


Présentation de Duccio et Sacha de l'utilisation de leur algorithme de shape matching à la Data Science Summer School. #DS3 #datascience #geomatics


Contact Us

4 rue Foucault, 75116 Paris
01 85 800 801

Advanced Aerial Imagery Analysis with Deep Neural Networks Explained in 5 Minutes.

There is no secret that when dealing with aerial images the best state-of-the-art results are achieved with deep learning models which come at the cost of their complexity. At the same time thanks to the Open Data, we can explore in a creative way even the most sophisticated techniques.

At nam.R we are working hard to build a Digital Twin of France, and to achieve that we use a lot of sources of information. One of the richest of them being aerial images. For us, humans, “reading” images is easy, but to teach computers how to deal with it is sometimes a real challenge (and fun!). In this post, we will show how we extract a rich description of buildings’ roofs from aerial images, particularly detecting their slopes. In Computer Vision jargon this task is called “object segmentation”, and to do that we chose to use a deep learning approach.

One of the current state-of-the-art segmentation models is the Mask R-CNN model published by researchers from Facebook. And we used this architecture implemented with a Keras framework.

To sum up, our deep learning model should be able to analyze the aerial images and detect roof slopes. This can help us understand the solar energy potential of the roof, and ultimately lead to a progress in nam.R’s vision: accelerating the ecological transition.

What Data Do We Need?

First of all, we need to define what kind of data is suitable for this task. We have a choice between satellite and aerial images. The main difference between them, in the context of our work, is the image resolution. Openly accessible satellite images have a resolution of several meters per pixel, while one can find aerial images with resolutions around 15-20 cm per pixel.

Because we would like to find some fine details on the images, roof slopes and ridges, we have gone with aerial images.

Training a deep learning model to detect roof slopes is a “supervised learning” task, so we need not only the images but also the labels of the slopes. So we created some labels ourselves. This is not a very exciting task, but it is a necessary steps to train a decent model.

This way we obtained two types of data to train the machine learning model: images of roofs and the labels for roof slopes.

A train data “image-label” pair looks like this:

To train a good model we need as much data as possible. Of course, we can label more roofs by hand, but it also possible to generate new samples just with the use of some simple transformations of original images and labels (“data augmentation”). This could be a rotation, a vertical or a horizontal flip and so on.

This way we can obtain a big enough dataset to train our deep learning model.

Deep Learning Model Which Fits Our Goal

Last few years, many high-performance deep neural networks were developed, and achieved impressive results on tasks of object detection. We chose Mask RCNN, a high-performance object segmentation network that was released in 2017. We adapted Matterport's implementation to be compatible with our aerial images and labels data source.

During the training, the model takes images and corresponding labels and learns its internal parameters to detect roof slopes on any new image. Because our dataset is quite small, a couple of hours of training already produce decent results.

One of the indicators that our model learned is the value of its loss function. The loss function is the criterion which the model tries to minimize, and it is usually an average of error between the real label and the predicted one.

Below are the loss function values for our model, which are constantly decreasing. The figures mean that the model learns well how to detect roof slopes:

Detect Roof Slopes on New Image

During the prediction phase, the model reads only aerial image and predicts the contours of the roof slopes in the image. Because the prediction step does not require complex calculations, it is possible, for example, to copy trained model to the production server and use it to analyze images in a real time.

We found roof slopes on the new image

We can see, that predicted labels for roof slopes are quite accurate, but, as always, there is some space for improvement. From the image above we can see how the roof slope detection doesn’t work well on roofs uncommon material such as metal.

One can imagine several ways to improve this model. For example, we can try to add more training samples for this type of roof material or do more data transformations to generate new samples.

But for the goal of our exploration of deep learning in advanced aerial image analysis, it is already a great result.

This post was just one example of a deep learning model used by nam.R to make France’s Digital Twin richer and closer to the reality. We will share our other techniques in future posts.

Stay tuned!

Developing a complex computer vision system, a case study : solar panels equipped roofs.

At nam.R we are working hard to build the Digital Twin of France. To this end, we use a lot of sources of information such as aerial images. Extracting useful information from the unstructured data that are images? Sounds like a job for the computer vision team !

To detect all solar panels on the roofs of french buildings, we used aerial imagery and the known outlines of the buildings, through a pipeline consisting of a solar panel outline detector and a filtering algorithm.

We took inspiration from the projects revolving around the state-of-the-art object segmentation deep learning algorithm known as Mask-RCNN. This algorithm is the newborn of a family of algorithms developed by Ross Girshick & al., in direct continuation of RCNN, Fast-RCNN and Faster-RCNN.

The chosen pipeline consists in two complementary parts :
- an object detector, more specifically an instance segmentation algorithm, meant to detect solar panels and extract their contours
- a filtering algorithm that takes all detections and filters out those that don’t match our business rules

While the filtering algorithm can easily be developed using the expert rules we chose to consider (size of the detected solar panels, their position regarding to the considered roof), the deep learning model depends directly of the data we will feed it with.

The first part of the project was, accordingly, to generate a dataset of roofs equipped with solar panels, and the matching labels. There are multiple existing tools for image annotations (VGG VIA, MIT LabelMe,...) that can be used as is. We chose the VGG Image Annotator. After a few (hundreds of) clicks, we ended up with a dataset we’re quite proud of.

Only then, we were able to train the Mask-RCNN model to detect solar panels.

The first version of our model wasn’t performing all that well and if was necessary to add more data to the training stage. We realized semi-supervised learning with automatic labelling.

This technique consists in using the model as a way to compute more labels that are then controlled by human operators and used as training data for a new, more robust version of the model. Controlling whether the proposed labels were right and correcting the wrong ones was way simpler than labelling by hand hundreds of images. Basically, we used our first model as a replacement for crowdsourcing !

After a few loops we fetched more data and matching labels and were able to train a model that had acceptable performances.

We transformed the raw output of the model into polygons in the same format and projection as our building polygons using geometric algorithms (Marching Squares, Douglas-Peucker) and geographic transformations. This enables us to directly filter out the potential false positives. We found out that roof windows, glass roofs and blue awning fabric were likely to be mistaken for solar panels due to their similar visual textures.

Fortunately, most false positives can be filtered out using the information of the building shape and position, but also expert rules concerning the minimal surfaces for solar panels.

An example of a false positive that was filtered out using its relative position to the building

The first conclusion we draw is that the combination of machine learning and expert rules can become a reliable framework, harnessing the power of machine learning algorithms and the robustness of business rules.

The second one was the use of our first imperfect model to help us label more data. Real data but synthetic labels, a great example of human-machine cooperation, isn’t it?

Finally, there are many different ways for computing the performances of this kind of pipelines. The deep learning model itself can be evaluated using metrics such as its mean average precision but we were mainly interested in the performances of the whole flow. Thus, we chose metrics that are less image-centric and oriented more towards information retrieval : precision, recall and overall accuracy. We added a geometric metric that indicates how well our predicted panels matched with the actual ones, the Intersection over Union (IoU).

Examples of the IoU computing : better overlaps mean better scores

We achieved the performances of 96% overall algorithm accuracy and 84% IoU on our test set, values we’re quite proud of.

The predictions of solar panels were integrated in nam.R’s Digital Twin and the information is already put to good use !