ELTA: An Enhancer against Long-Tail for Aesthetics-oriented Models

Authors: Limin Liu, Shuai He, Anlong Ming, Rui Xie, Huadong Ma

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on four representative datasets (AVA, AADB, TAD66K, and PARA) show that our proposed ELTA achieves state-of-the-art performance by effectively mitigating the long-tailed issue in IAA datasets.
Researcher Affiliation Academia 1School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing, China. Correspondence to: Anlong Ming <mal@bupt.edu.cn>.
Pseudocode No The paper describes the steps of its proposed methods (e.g., TFA, FLSA, APDS) but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks or figures.
Open Source Code Yes All resources are available in here.
Open Datasets Yes Datasets. We evaluate the performance of our approach on four representative datasets: AVA (Murray et al., 2012), AADB (Kong et al., 2016), TAD66K (He et al., 2022), PARA (Yang et al., 2022)
Dataset Splits Yes AADB is an aesthetic attribute dataset containing about 10,000 images, each scored from 1 to 5. Following the previous work (Kong et al., 2016; Zhu et al., 2020; 2021), we use the standard split with 8,500 for training, 1,000 for testing and 500 for validation.
Hardware Specification No The paper states, 'Therefore, we select the Swin Transformer V2 (Liu et al., 2022), known for its versatility and popularity, as our network backbone.' However, it does not specify any particular hardware, such as GPU models, CPU types, or memory, used for conducting the experiments.
Software Dependencies No The paper mentions using the 'Adam optimizer' and 'Swin Transformer V2' as the backbone, but it does not specify version numbers for any software dependencies like programming languages (e.g., Python), deep learning frameworks (e.g., PyTorch, TensorFlow), or specific libraries.
Experiment Setup Yes The training process is optimized using the Adam optimizer, with a batch size of 48.