Anticipating Future Relations via Graph Growing for Action Prediction
Authors: Xinxiao Wu, Jianwei Zhao, Ruiqi Wang2952-2960
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on two action video datasets demonstrate the effectiveness of our method. and sections like Experiments Datasets, Comparison with State-of-the-art Methods, Ablation Studies. |
| Researcher Affiliation | Academia | 1Beijing Laboratory of Intelligent Information Technology School of Computer Science, Beijing Institute of Technology, Beijing, China 2School of Information Science and Technology, Northeast Normal University, Changchun, China {wuxinxiao,wang ruiqi}@bit.edu.cn, zhaojw374@nenu.edu.cn |
| Pseudocode | No | No pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating that the source code for the methodology is openly available. |
| Open Datasets | Yes | We conduct experiments to evaluate our method on two datasets: UCF101 (Soomro, Zamir, and Shah 2012) and 20BN-something-something (Goyal et al. 2017). |
| Dataset Splits | Yes | The UCF101 dataset... provides three official splits for training and validation and we report the average accuracy over the three splits by following the standard practice. and The 20BN-something-something dataset... There are 11,101 short videos for training and 1,568 videos for validation. and We randomly split 10% of the training set as a validation set. |
| Hardware Specification | Yes | We use Py Torch 0.4.1 and train the model for 500 epochs on one GTX-1080Ti GPU. |
| Software Dependencies | Yes | We use Py Torch 0.4.1 and train the model for 500 epochs on one GTX-1080Ti GPU. |
| Experiment Setup | Yes | In the relation reasoning, the unit number of GRU layers is set to 512 and the number of propagation is set to 3. The feature dimensions of both initial nodes and edges are reduced to 512 by a linear layer. In the relation synthesizing, the dimension of graph representation feature is set to 512. We use two graph convolutional layers to rebuild the adjacency matrix Asyn and one graph convolutional layer to refresh the node feature F. The number of scale is set to 5 for 20BN-something-something and 8 for UCF101. The balance parameter λ1 is set to 0.125 and λ2 is set to 1. We randomly split 10% of the training set as a validation set. All the networks are trained from scratch with an initial learning rate of 0.00005. The Adam optimizer (Kingma and Ba 2015) and the SGD optimizer are employed with a batch size of 48 for optimization. We use Py Torch 0.4.1 and train the model for 500 epochs on one GTX-1080Ti GPU. |