Learning Visual Context for Group Activity Recognition
Authors: Hangjie Yuan, Dong Ni3261-3269
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Systematic experiments demonstrate each module s effectiveness on either branch. Visualizations indicate that visual contextual cues can be aggregated globally by TCE. Moreover, our method achieves state-of-the-art results on two widely used benchmarks using only RGB images as input and 2D backbones. |
| Researcher Affiliation | Academia | 1 College of Control Science and Engineering, Zhejiang University, Hangzhou, China 2 State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou, China |
| Pseudocode | No | No pseudocode or clearly labeled algorithm blocks were provided in the paper. |
| Open Source Code | No | The paper does not contain any statements about releasing source code or provide links to a code repository. |
| Open Datasets | Yes | There are two frequently adopted datasets named the Volleyball dataset (VD) and the Collective Activity dataset (CAD). The Volleyball dataset (Ibrahim et al. 2016) gathers from 55 video recordings of volleyball games, which are clipped and split into 3493 training clips and 1337 testing clips. The Collective Activity dataset (Choi, Shahid, and Savarese 2009) composes of 44 clips... |
| Dataset Splits | No | The paper states, "The Volleyball dataset (...) is split into 3493 training clips and 1337 testing clips." and "The train set and test set split follows (Qi et al. 2018)" for CAD. However, it does not explicitly provide details about a separate validation split, its size, or how it was created for either dataset. |
| Hardware Specification | No | The paper does not specify any particular hardware used for experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions using "Adam optimizer" but does not specify version numbers for any software dependencies like deep learning frameworks (e.g., PyTorch, TensorFlow) or programming languages. |
| Experiment Setup | Yes | For the training of VD, we adopt the Adam optimizer (Kingma and Ba 2014) with its hyper-parameter fixed to β1 = 0.9, β2 = 0.999 and ϵ = 10 8. We use a minibatch size of 6 and train the network in 160 epochs with an initial learning rate 10 4, which decreases by a factor of 2 every 40 epochs. For TCE module, we set the encoding dimension to dc = 128 and dropout ratio to 0.1. For STBi P module, we set the threshold for the gate function to θ = 0.3H and set dw = dy. For the training loss, we use λ = 1 for all experiments. |