The long-tailed data distribution poses an enormous challenge for training neural networks in classification.A classification network can be decoupled into a feature extractor and a *** paper takes a semi-discrete opt...
详细信息
The long-tailed data distribution poses an enormous challenge for training neural networks in classification.A classification network can be decoupled into a feature extractor and a *** paper takes a semi-discrete opti-mal transport(OT)perspective to analyze the long-tailed classification problem,where the feature space is viewed as a continuous source domain,and the classifier weights are viewed as a discrete target *** classifier is indeed to find a cell decomposition of the feature space with each cell corresponding to one *** imbalanced training set causes the more frequent classes to have larger volume cells,which means that the classifier's decision boundary is biased towards less frequent classes,resulting in reduced classification performance in the inference ***,we propose a novel OT-dynamic softmax loss,which dynamically adjusts the decision boundary in the training phase to avoid overfitting in the tail *** addition,our method incorporates the supervised contrastive loss so that the feature space can satisfy the uniform distribution *** and comprehensive experiments demonstrate that our method achieves state-of-the-art performance on multiple long-tailed recognition benchmarks,including CIFAR-LT,ImageNet-LT,iNaturalist 2018,and Places-LT.
Semantic segmentation is an important task in the field of computer vision and is widely used in fields such as medical image analysis and autonomous driving. However, when the recognized subject and the background ar...
详细信息
With the development of 5G technology, the Internet of things and the Internet, the amount of data mastered by people is rapidly increasing. How to store and query these data becomes more and more important. InterPlan...
详细信息
Globally, cancer remains a leading cause of death, affecting millions of people each year. Accurate medical imaging is crucial for the effective planning of radiotherapy. However, repeated exposure to radiation from C...
详细信息
To handle problems of data information recognition in various industries, the issue of handwritten digit recognition is solved in this paper by implementing four machine learning methods for the task of classification...
详细信息
Combining fluid mechanics and traffic flow theory, this paper proposes a new macro-micro integrated simulation method, which regards vehicles driving in the city as fluid, and calculates the density and speed of traff...
详细信息
In recent years, traditional deep learning has been widely used in the time series prediction of air quality, but this kind of model has many shortcomings in the input selection of meteorological related data. Based o...
详细信息
The methodologies based on neural networks are substantial to accomplish sentiment analysis in the Social Internet of Things (SIoT). With social media sentiment analysis, significant insights can produce efficient and...
详细信息
Industrial tasks with traditional cloud architecture have many disadvantages, which are mainly longer delay of data transmission and difficulty in software deployment. In view of the above disadvantages, this paper us...
详细信息
Regarding the current existing modeling methods, procedural modeling is a fast and convenient method. The more perfect shape in use now is the CGA shape, which is a new shape syntax for procedural modeling of CG archi...
详细信息
暂无评论