Search Results for author:
Found 4 papers, 0 papers with code
An improved helmet detection method for YOLOv3 on an unbalanced dataset
The YOLOv3 target detection algorithm is widely used in industry due to its high speed and high accuracy, but it has some limitations, such as the accuracy degradation of unbalanced datasets.
Machine learning approach of Japanese composition scoring and writing aided system's design
Though it is still difficult for machines to directly evaluate a composition at the semantic and pragmatic levels, especially for Japanese, Chinese and other language in high context cultures, we can make machine evaluate a passage in word and grammar levels, which can as an assistance of composition rater or language learner.
Approaches of large-scale images recognition with more than 50,000 categoris
Though current CV models have been able to achieve high levels of accuracy on small-scale images classification dataset with hundreds or thousands of categories, many models become infeasible in computational or space consumption when it comes to large-scale dataset with more than 50, 000 categories.
Spine Landmark Localization with combining of Heatmap Regression and Direct Coordinate Regression
Landmark Localization plays a very important role in processing medical images as well as in disease identification.
