Vincent Liu

May 29, 2024 | ai.gopubby.com | Vincent Liu

By Vincent Liu Figure 1. Plastic surface samples. Source: image by author. Anomaly detection in computer vision has come a long way. A variety of anomaly detection works have already been utilized in real applications. The unsupervised learning solutions in anomaly detection have been beneficial to productization. We can train a model to identify subtle differences in images by solely using normal samples...

Mar 30, 2024 | medium.com | Vincent Liu

After 46 and a half hours with Banishers: Ghosts of New Eden, I find myself with a familiar hollow feeling as the credits rolled, that feeling when you suddenly come to the end of something and realise that there is no more left. I’ve come to recognise it as a sign of an excellent game. In hindsight, I should have expected that from a game developed by DON’T NOD, famous for their excellently written and delivered stories.

Mar 29, 2024 | medium.com | Vincent Liu

Albumentation v1.4.2What is MixupMixup is a data augmentation technique that synthesize a new sample by combining two existing data points. The following formula comes from the original paper where x and y represent the data value and corresponding label. The parameter x̂ is the linear combination of the two original data points xᵢ and xⱼ with the weight λ. Similarly, the parameter ŷ is combined label in any format resulted from the two original labels yᵢ and yⱼ .

Mar 25, 2024 | medium.com | Vincent Liu

For people who are interested in anomaly detection, Anomalib is a useful tool that is highly recommended to keep it on hands. It is constantly updated with state-of-the-art works so one can quickly access the capability of the top-notch models. In this article, we would like to share an exercise based on the latest Anomalib library v1.0.0. The following sections covers the development environment, dataset preparation, training/testing script and the final evaluation result.

Jan 12, 2024 | levelup.gitconnected.com | Vincent Liu

how criminal-legal data can be valuable yet dangerousThe past few years have witnessed a rapid increase in the interest of using data in fields from those that are historically data-heavy, such as healthcare and finance to those that are not, law enforcement as an example. The tradition of crime field is associated with qualitative tools like interviews and focus groups, and human-based judgment, calling from experiences and observations of real-time dynamics.