Facebook AI has unveiled UVO (Unidentified Video Objects), a new dataset to boost the Artificial Intelligence (AI) research on open-world segmentation. The dataset will boost AI vision, thus helping machines identify unfamiliar objects with a similar ease and precision as humans thanks to the vastness of the dataset and the quality of the video content.

In the past few years, object segmentation has risen in popularity as a research subject to augment computer vision as it is a key to develop a machine's ability to identify objects and their correct location.

In the past, researchers have innovated approaches such as Mask R-CNN and MaskProp. However, these methods have had limited success because these algorithms already assume that everything that they see has been already preordained for detection and segmentation beforehand. There are innumerable object concepts that the AI models have never seen or learned in real-world applications, such as embodied AI or augmented reality assistants.

However, this is not the case with humans. Humans can detect unfamiliar objects without having prior knowledge about them. The Facebook AI team set out to resolve this question. Their research has led to surprising results such as machines detecting and segmenting any object they encounter regardless if it’s previously known or unknown.

"Teaching machines to detect any object — whether familiar or entirely novel — will enable them to perform a wide range of important tasks that are beyond the abilities of today’s AI. Object search, instance registration, human-object interaction modeling, and human activity understanding all require open-world prediction abilities, for example," write the researchers in an Facebook official blog.

Facebook's new dataset contains an exhaustive collection of high-quality object mask annotations that contains real-world videos of action recognition benchmarks. These videos have on an average, 13.5 unique object instances, by far the highest than any existing datasets.

Facebook AI researchers believe UVO is a versatile test bed for developing novel approaches to open-world segmentation while inspiring more research on building comprehensive understanding outside just classification or detection.