Extensive experience at managing team of several people between 10 and 30 members using many useful technology like Slack, Git, Bitbucket and MantisBT.
I created South Valley Roleplay, an online game become one of the most important servers of San Andreas Multiplayer in Italy, and I managed it until 2017 dealing with staff recruitment, community and account management and projecting new features. We reached thousands of subscribers and players in the years.
In 2017 I launched FantaVikings and I created a team of about 15 people to manage it and make it grow. In less than one month we reached the 500 subscribers and we were interviewed by well-known online italian sites.ARTICLE INTERVIEW
In 2017 I conducted an agreement to unite the three main San Andreas Multiplayer servers in Italy and I've managed the new server guiding a large team of almost 30 people and a very heterogeneous community for months. It was the last really big project of this mod in the italian panorama.
Identify the perceptive differences between two images is a big problem of colorimetry industry. I studied the evolution of images changing resolutions and I used colorimetry instruments and neural networks to understand the minimum resolution needed to identify differences between two images. In this example we can see two images at 720p and 4K resolutions. If you look at the images they appear very similar but if you try to concentrate to a particular detailed zone, such as the leaves, you can notify that in the image at 4K they looks like more detailed and in the one at 720p they appear more noisy. We could think that we should work in high resolutions to identify differences between images because we can see more details. Is it true?
To emulate the visual perception of the human being, it was decided to train a neural network. To do this I started from a dataset of 40 images at a very high resolution of 4K, in the L*a*b* color space. For each of these, 5 were generated at minor or equal resolutions, to which 3 color filters were applied to 3 different intensities. Each of these images has been divided into 9 areas of equal size and for each of these numerical values descriptive of the characteristics have been calculated. These values were used by the network to understand the content of the images to be compared. 1800 comparisons were then made between the original images and the respective filtrates, assigning to each pair an index of similarity from L, low similarity, to H, high similarity. In order to maximize the accuracy of the neural network, 100 were trained and the 5 most suitable were selected. Taking advantage of these networks, an ensemble classifier was created, which is a complex of neural networks in which the final result is the one repeatedly extracted from the individual networks. It was therefore possible to obtain an accuracy of just under 90%.
The network then tried to judge pairs of images at all resolutions taken into consideration and the opinion given by the network is almost always unchanged and rarely tends to decrease, thus highlighting a slight accentuation of the differences at the end of the resolution. This trend was also confirmed by analyzing the opinions of the human observer. In 94% of cases, in fact, the judgment remains unchanged or worsens, as the resolution decreases. This discovery could allow the colorimetry industry to save 97% on the costs and times of this process, allowing it to perform faster, more precise and numerous controls and thus improve their production processes.
MY ADDRESSContact me and let's start to collaborate together on a new thrilling project!