Spatial-temporal Causal Inference for Partial Image-to-video Adaptation
Authors: Jin Chen, Xinxiao Wu, Yao Hu, Jiebo Luo1027-1035
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on several video datasets have validated the effectiveness of our proposed method. Table 1 shows the classification accuracies of different methods on both S U and E H tasks. To better understand the effect of each component, we conduct ablation experiments on both S U and E H tasks, as shown in Table 2. |
| Researcher Affiliation | Collaboration | Jin Chen,1 Xinxiao Wu,1 Yao Hu,2 Jiebo Luo3 1Beijing Laboratory of Intelligent Information Technology School of Computer Science, Beijing Institute of Technology, Beijing, China 2Alibaba Youku Cognitive and Intelligent Lab 3Department of Computer Science, University of Rochester, Rochester NY 14627, USA {chen jin, wuxinxiao}@bit.edu.cn, yaoohu@alibaba-inc.com, jluo@cs.rochester.edu |
| Pseudocode | No | The paper describes methods and processes but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/Chen Jin BIT/HPDA. |
| Open Datasets | Yes | We conduct experiments on two video benchmarks, i.e., UCF101 (U) (Soomro, Zamir, and Shah 2012) and HMDB51 (H) (Kuehne et al. 2011). With the UCF101 dataset as the target domain, we use the Standford40 (S) dataset (Yao et al. 2011) as the source domain. With the HMDB51 dataset as the target domain, we use the EADs (E) dataset (Yu et al. 2018) as the source domain. |
| Dataset Splits | No | The paper states which parts of the datasets are used as source and target domains, and the number of shared classes, but does not provide specific training/validation/test dataset splits with percentages or sample counts for reproducibility beyond that. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments. It only discusses the general architecture and pre-trained models. |
| Software Dependencies | No | The paper mentions software components like ResNet-50, I3D networks, and the Adam solver, but it does not specify version numbers for any software, libraries, or frameworks used for implementation (e.g., Python, PyTorch, TensorFlow). |
| Experiment Setup | Yes | The paper provides specific experimental setup details including: 'The Adam solver (Kingma and Ba 2015) with the batch size of 16, including 8 source images and 8 target videos. The trade-off parameter λ in Eq. (9) is set to 100. The dimension of the noise z is set to 1024. We set the threshold τ in Eq. (7) as 0.5... All the networks are trained from scratch with 400 epochs. We keep the same learning rate for the first 200 epochs and linearly decay the rate to zero for the next 200 epochs. The initial learning rate of the S U and E H tasks are set to 0.0001 and 0.00005, respectively.' |