UniTTA
收藏资源简介:
UniTTA数据集由清华大学创建,是一个综合性的测试时间适应性(TTA)评估基准,涵盖了36种不同的场景。该数据集通过马尔可夫状态转移矩阵生成,考虑了域和类的独立维度,以及平衡/不平衡和i.i.d./非i.i.d./连续条件。数据集的创建过程结合了时间和不平衡因素,通过调整马尔可夫矩阵的参数来模拟真实世界的数据流。UniTTA数据集主要应用于测试时间适应性研究,旨在解决模型在实际部署中因测试数据域偏移导致的性能下降问题。
The UniTTA dataset, developed by Tsinghua University, is a comprehensive test-time adaptation (TTA) evaluation benchmark spanning 36 distinct scenarios. Generated using Markov state transition matrices, this dataset accounts for independent dimensions of domain and category, alongside balanced/unbalanced, i.i.d./non-i.i.d., and continuous conditions. Its construction process integrates temporal and imbalance factors, simulating real-world data streams by adjusting the parameters of the Markov matrices. Primarily utilized in test-time adaptation research, the UniTTA dataset aims to resolve the performance degradation of models induced by domain shift in test data during real-world deployment.
UniTTA 数据集
- 数据集名称: UniTTA
- 代码状态: 即将发布

- 1UniTTA: Unified Benchmark and Versatile Framework Towards Realistic Test-Time Adaptation清华大学 · 2024年



