Structured and Sparse Annotations for Image Emotion Distribution Learning
Authors: Haitao Xiong, Hongfu Liu, Bineng Zhong, Yun Fu363-370
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiment results demonstrate that our proposed SSDL significantly outperforms the state-of-the-art methods. |
| Researcher Affiliation | Academia | Haitao Xiong,1 Hongfu Liu,2 Bineng Zhong,3 Yun Fu4 1School of Computer and Information Engineering, Beijing Technology and Business University, Beijing, China 2Michtom School of Computer Science, Brandeis University, Waltham, USA 3chool of Computer Science and Technology, Huaqiao University, Xiamen, China 4Department of Electrical and Computer Engineering, Northeastern University, Boston, USA |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing open-source code for the described methodology or a link to a code repository. |
| Open Datasets | Yes | To make this comparison, three image emotion distribution datasets are selected, including Emotion6 (Levi and Hassner 2015), Flickr LDL and Twitter LDL (Yang, Sun, and Sun 2017). Emotion6 is widely used as a benchmark dataset for emotion classification, which contains 1,980 images collected from Flickr. Flickr LDL and Twitter LDL are two datasets collected mainly for emotion distribution learning. Flickr LDL contains 11,150 images and Twitter LDL contains 10,045 images. |
| Dataset Splits | Yes | All datasets are randomly split into 75% training, 20% testing and 5% validation sets. The validation set is used for choosing the best parameters of our methods. |
| Hardware Specification | Yes | All our experiments are carried out on a NVIDIA GTX Titan Xp GPU with 12GB memory. |
| Software Dependencies | No | The paper mentions using 'VGG-19 architecture' but does not specify any software dependencies with version numbers. |
| Experiment Setup | Yes | The learning rates of the convolution layers, the first two fullyconnected layers and the classification layer are initialized as 0.001, 0.001 and 0.01. We fine-tune all layers by back propagation through the whole net using mini-batches of 32 and the total number of epochs is 20 for learning. |