What and Where the Themes Dominate in Image
Authors: Xinyu Xiao, Lingfeng Wang, Shiming Xiang, Chunhong Pan9021-9029
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the COCO and Flickr30K datasets achieve superior results when compared with the state-of-the-art methods. |
| Researcher Affiliation | Academia | Xinyu Xiao,1,2 Lingfeng Wang,1 Shiming Xiang,1,2 Chunhong Pan1 1National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences 2School of Artificial Intelligence, University of Chinese Academy of Sciences |
| Pseudocode | No | The paper includes mathematical formulas and descriptions of models but does not present them in a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | The paper does not provide an explicit statement about releasing source code or a link to a code repository for the methodology described. |
| Open Datasets | Yes | We evaluate our method on two widely used datasets. The first one is Flickr30K (Young et al. 2014)... The other more challenging dataset COCO (Lin et al. 2014)... |
| Dataset Splits | Yes | Following the publicly splits1, we divide 29,014, 1,000 and 1,000 images for training, validation and testing, respectively. |
| Hardware Specification | No | The paper does not specify the exact hardware used for the experiments, such as specific GPU or CPU models. |
| Software Dependencies | No | All the experiments are implemented with Pytorch (Paszke et al. 2017). (No version number for Pytorch is specified). |
| Experiment Setup | Yes | We adopt the Res Net152 model (He et al. 2016)... The dimensions of the visual feature channel and all the LSTM hidden states are set to the same length as the concept vector. And the length of the theme vector is determined by the complexity of the dataset, which is set to 1,000, 200 for COCO and Flickr30k, respectively. The proposed RL-based method is applied to optimize the just MLE trained model with the CIDEr metric, and λθ is set to 2. At each epoch, the validation set is used to evaluate the model, and the model with the best CIDEr score is selected for testing. |