It's been a while since the last meeting, more than a year, but new forces have joined the group and are eager to reignite the idea: making Rome a great place for software engineering and data science.
Location: Immobiliare Labs, Via di Santa Prassede 24, Rome - Italy
Date: May 8th 2024
Schedule:
- 18:30 ๐ช Door Opening
- 19:00 ๐ค Talk 1 - Computer Vision IRL: from idea to on-premise deployment, by Egon Ferri, computer vision engineer @ Immobiliare.it
- 19.30 ๐ค Talk 2 - Geospatial Data Integration with Jupyter Notebook and OS software, by Luigi Selmi, scientific software developer @ datiaperti.it
- 20.30 ๐ค Socializing
The presentations should be of interest to anyone working on satellite images, geospatial data, machine learning, and software engineering. As always the group is open to anyone willing to present her or his work for future meetings.
Here is a short description of the two presentations:
1. Computer Vision IRL: from the idea to the on-premise deployment
In the field of Computer Vision (CV), attention is often focused on the development of deep learning models, neglecting crucial stages of problem definition such as data preparation, model deployment, and subsequent monitoring.
In this talk, we will illustrate our approach to the development of CV services through a real use case: the classification of a property's environments using exclusively open-source technologies. Specifically, we will discuss:
- How we transition from business needs to technical requirements for the model and service;
- The process of creating the image dataset and developing and optimizing the model's performance;
- The deployment in an on-premise environment of the CV service using a combination of a FastAPI-branded API proxy and TorchServe-based model serving to ensure a scalable service that guarantees high throughput and low latency;
- How we implemented our monitoring system to ensure the efficiency and correctness of the service, overseeing the hardware, data quality, model accuracy, and business KPIs.
Through this approach, we have created a high-performing CV service, demonstrating the effectiveness of open-source solutions in meeting complex business needs.
2. Geospatial Data Integration with Jupyter Notebook and OS software
In many applications of geoscience, for example in emergency management or risk assessment, one wants to know the spatial context of the area in which a certain event occurred and link the spatial information to external data sources through a named entity. For example, it is not enough to know the extent of a flooded area, we want to know in which administrative area it occurred. In other words, we want to know the spatial relation between the area in which an event occurred and the area that belongs to or is run by an authority. In this notebook we will see how such spatial relations are implemented in GeoPandas. We will deal only with topological relations, a subclass of spatial relations that don't need a metric, that is a way to compute the distance. We will see how a Jupyter Notebook can be used to perform a data analysis starting with some simple spatial objects, points and polygons, to see what kind of topological relations can be established between them and finally we will apply those topological relations between layers in a real-world example to merge a dataset of landslides and the administrative boundaries of the Marche region, to extract some useful statistics.
Sign Up!
Space is limited, so RSVP today to secure your spot!
Please note:
Remember to RSVP using your full name for security reasons.
If you can't attend, please let us know by 15/04 so we can plan accordingly. We look forward to seeing you there! ๐