Accurate Temporal Action Proposal Generation with Relation-Aware Pyramid Network
Authors: Jialin Gao, Zhixiang Shi, Guanshuo Wang, Jiani Li, Yufeng Yuan, Shiming Ge, Xi Zhou10810-10817
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the challenging Activity Net and THUMOS14 benchmarks demonstrate our Rap Net generates superior accurate proposals over the existing state-of-the-art methods. |
| Researcher Affiliation | Collaboration | Jialin Gao,1 Zhixiang Shi,2 Guanshuo Wang,1 Jiani Li,2 Yufeng Yuan,2 Shiming Ge,3 Xi Zhou1,2 1Cooperative Medianet Innovation Center, Shanghai Jiao Tong University 2Cloud Walk Technology Co., Ltd, China 3Institute of Information Engineering, Chinese Academy of Sciences |
| Pseudocode | No | The paper describes algorithms conceptually and mathematically but does not include a dedicated pseudocode block or algorithm listing. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository. |
| Open Datasets | Yes | THUMOS 2014. This dataset consists of 13320 trimmed videos of 101 categories from UCF-101 for training, 1010 and 1574 untrimmed videos for validation and test set respectively. Activity Net-v1.3. It contains 19994 videos labeled in 200 classes. |
| Dataset Splits | Yes | THUMOS 2014. This dataset consists of 13320 trimmed videos of 101 categories from UCF-101 for training, 1010 and 1574 untrimmed videos for validation and test set respectively. Activity Net-v1.3. It contains 19994 videos labeled in 200 classes. it is divided into training, validation and test with a ratio of 0.5, 0.25, 0.25 respectively. |
| Hardware Specification | No | The paper mentions using a 'Res Net-50 C3D' model and 'C3D features' but does not specify any hardware details like GPU models, CPU types, or memory used for the experiments. |
| Software Dependencies | No | The paper mentions various components like 'Res Net-50 C3D', 'K-means algorithm', 'YOLO', and 'Soft-NMS' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | We train our Rap Net with batch size of 16 and initialize learning rate as 0.005 with Cosine Decline Learning Strategy for 18 epochs, which is warmed up four epochs in a linear growth mode. We set N = 6, M = 2 for almost experiments. We empirically set λ1 = 10λ2, λ3 = 0.0005 for experiments. A negative instance will be ignored if the highest Io U overlap between groundtruth instances with all proposal predictions is larger than a threshold θiou, empirically 0.5. For a ground truth instance φ = [ts, te], the label of each temporal location lying in the expanded region [ts dη, te + dη] will be set to 1, where d = te ts. We set λconf = 0.2 and other weights as 1 for training Rap Net. |