Alibaba’s new machine vision algorithm, AliceMind, distinguishes images with an accuracy of 81.26%, compared to 80.83% for humans.
The Visual Question Answering (VQA) Challenge 2021 revealed the best machine vision algorithm – AliceMind.
During the competition, more than 250 thousand images were used, which the algorithms studied, and then had to answer 1.1 million text questions about what they saw.
Alibaba’s AliceMind algorithm did the best: it even surpassed humans in accuracy. Over time, this gap will only widen, the developers say.
Computer vision is one of the most active research and development areas in China. It can be used to search for products on websites, analyze medical images, initially diagnose diseases, and develop self-driving cars.
Alibaba believes that the algorithm will not replace human labor, but it will be able to remove the routine from it. For example, computer vision will be able to conduct primary medical analysis without distracting doctors from their main responsibility – to make an accurate diagnosis.