Computer-Vision-Dashboard für Unternehmen: Medien hochladen, Erkennung konfigurieren, Ergebnisse bewerten und Kennzahlen direkt prüfen.
ArmoredEye is a project designed to improve the detection of destroyed military vehicles. The goal is not only to distinguish between tanks and combat aircraft, but also to determine the exact model designation. In the long term, the aim is to identify the origin and model so that our soldiers on the battlefield receive precise information that gives them a decisive advantage.
On February 24, 2022, Russia decided to declare war on Ukraine, regardless of the cost in human lives—men, women, and children. The “ArmoredEye” project was launched to advance Europe and forward-looking technologies. It uses web scraping methods and YOLO models to demonstrate that, with minimal effort and a touch of creativity, technologies such as computer vision models are now a realistic possibility. But more on that later.
On February 24, 2022, Russia decided to declare war on Ukraine, regardless of the cost in human lives—men, women, and children. The “ArmoredEye” project was launched to advance Europe and forward-looking technologies. It uses web scraping methods and YOLO models to demonstrate that, with minimal effort and a touch of creativity, technologies such as computer vision models are now a realistic possibility. But more on that later.
Training a computer vision model requires an incredible amount of reliable and robust data. This goes hand in hand with computational power, which is extremely costly at this scale. To minimize costs, the YOLO model was specifically selected, with YOLO-11n and YOLO-11s being used. Since YOLO is open source, it is easy to continue using the models with their weights and biases and adapt them to specific needs. After several trials, YOLO delivered the most reliable results compared to the other models. The models were able to generalize very well, which was advantageous in the case of military vehicles, as they closely resemble civilian vehicles or aircraft.
Training a computer vision model requires an incredible amount of reliable and robust data. This goes hand in hand with computational power, which is extremely costly at this scale. To minimize costs, the YOLO model was specifically selected, with YOLO-11n and YOLO-11s being used. Since YOLO is open source, it is easy to continue using the models with their weights and biases and adapt them to specific needs. After several trials, YOLO delivered the most reliable results compared to the other models. The models were able to generalize very well, which was advantageous in the case of military vehicles, as they closely resemble civilian vehicles or aircraft.
Dabei wurden folgende YOLO Modelle gestestet. YOLO11n, YOLO11s, YOLO11m und YOLO11l Dabei wurde sehr schnell festgestellt das YOLO11m und YOLO11l umfassende ergebnisse erzielt wurden jedoch die Rahmen des Budgest gesprangt haben. Das Ziel bleibt weiterhin Modelle schenell auf bedrafs fall anzupassen.
Placeholder: Note why you chose YOLO, Flask, dashboard controls, and the metrics that are shown in the interface.
Placeholder: Write what worked well, what was difficult, and what you would improve in a next version of ArmoredEye.