Retrieving and Classifying Affective Images via Deep Metric Learning
Authors: Jufeng Yang, Dongyu She, Yu-Kun Lai, Ming-Hsuan Yang
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive evaluations on four widely-used affective datasets, i.e., Flickr and Instagram, IAPSa, Art Photo, and Abstract Paintings, demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods on both affective image retrieval and classification tasks. |
| Researcher Affiliation | Academia | Jufeng Yang,1 Dongyu She,1 Yu-Kun Lai,2 Ming-Hsuan Yang3 1College of Computer and Control Engineering, Nankai University, Tianjin, China 2School of Computer Science and Informatics, Cardiff University, Cardiff, United Kingdom 3School of Engineering, University of California, Merced, USA |
| Pseudocode | No | The paper describes the proposed algorithm in prose sections and figures but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating the public availability of its source code. |
| Open Datasets | Yes | We perform our experiments on four datasets, including Flickr and Instagram (FI) (You et al. 2016), IAPSa, Art Photo and Abstract Paintings (Machajdik and Hanbury 2010). FI is collected from social websites by querying with Mikels eight emotions as keywords. ... The International Affective Picture System (IAPS) (Lang, Bradley, and Cuthbert 2008) is a common stimulus dataset which is widely used in visual emotion understanding research, from which IAPSa selects 395 pictures annotated with the same eight emotion categories. |
| Dataset Splits | Yes | First, the network is initialized with the weights trained for the large-scale dataset, and then fine-tuned on the FI dataset with the FC8 layer changed to 8, which is split randomly into 80% training, 5% validation and 15% testing set. ... With the help of transfer learning, we also employ our framework on three datasets with limited training examples. In details, we transfer the parameters of the network fine-tuned on the FI as well as the hyper-parameters to the small-scale datasets, which are split into 80% training and 20% testing set randomly. We conduct 5-fold validation and report the average performance. |
| Hardware Specification | Yes | All our experiments are carried out on two NVIDIA GTX 1080 GPUs with 32 GB CPU memory on-board. |
| Software Dependencies | No | The paper mentions building the framework based on 'Google Net Inception' and using 'LIBSVM' but does not specify version numbers for these or other ancillary software dependencies like programming languages or deep learning frameworks. |
| Experiment Setup | Yes | The learning rates of the convolutional layers and the last fully-connected layer are initialized as 10 4 and 10 3, respectively. We fine-tune all layers by stochastic gradient descent (SGD) through the whole net using batches of 128... A total of 100 epochs are run... We set the margin α in the triplet loss to 0.2, while α1 and α2 in the sentiment loss are set to 0.2, 0.1, respectively. We set the weight ω as 0.7... |