Revisiting Image Aesthetic Assessment via Self-Supervised Feature Learning
Authors: Kekai Sheng, Weiming Dong, Menglei Chai, Guohui Wang, Peng Zhou, Feiyue Huang, Bao-Gang Hu, Rongrong Ji, Chongyang Ma5709-5716
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive quantitative experiments on three benchmark datasets and demonstrate that our approach can faithfully extract aesthetics-aware features and outperform alternative pretext schemes. Moreover, we achieve comparable results to state-of-the-art supervised methods that use 10 million labels from Image Net. |
| Researcher Affiliation | Collaboration | Kekai Sheng,1,2 Weiming Dong,1 Menglei Chai,3 Guohui Wang,3 Peng Zhou,4 Feiyue Huang,2 Bao-Gang Hu,1 Rongrong Ji,5 Chongyang Ma6 1Institute of Automation, Chinese Academy of Sciences, 2Youtu Lab, Tencent, 3Snap Inc. 4North China Electric Power University, 5Xiamen University, China, 6Kuaishou Technology |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions using pre-trained models from other authors (e.g., 'In our experiments, we use the pre-trained models released by the authors for reliable and fair comparisons'), but does not provide a statement or link for the open-source code of the methodology described in this paper. |
| Open Datasets | Yes | Aesthetic Visual Analysis (AVA). The AVA dataset (Murray, Marchesotti, and Perronnin 2012) contains approximately 250, 000 images. Aesthetics with Attributes Database (AADB). The AADB dataset (Kong et al. 2016) contains 10, 000 images with aesthetic ratings. Chinese University of Hong Kong-Photo Quality Dataset (CUHK-PQ). The CUHK-PQ dataset (Luo, Wang, and Tang 2011) contains 17, 690 images with binary aesthetic labels. |
| Dataset Splits | Yes | Aesthetic Visual Analysis (AVA). ...230, 000 images for training and 20, 000 for testing. Aesthetics with Attributes Database (AADB). ...8500, 500, and 1000 images for training, validation, and testing, respectively. Chinese University of Hong Kong-Photo Quality Dataset (CUHK-PQ). ...a random 50/50 split and a five-fold split for cross-validation. We use the former one in our experiments. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'SGD optimization' and 'Nesterov momentum' but does not specify any software libraries with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | We apply SGD optimization using a batch size of 64, with the Nesterov momentum of 0.9 and the weight decay of 5e 4. We begin with a learning rate of 0.1, dropped it by a factor of 0.2 after every 10 epochs. To eschew training oscillating, we activate Ltrp(p, t) with λ of 0.02 after the first 30 epochs. The following adaptation stage shares the same settings except that the learning-rate starts from 0.01. |