Exposing the Self-Supervised Space-Time Correspondence Learning via Graph Kernels
Authors: Zheyun Qin, Xiankai Lu, Xiushan Nie, Yilong Yin, Jianbing Shen
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Video Hi Graph achieves superior performance and demonstrates its robustness across the benchmark of label propagation tasks involving objects, semantic parts, keypoints, and instances. Our algorithm implementations have been made publicly available at https://github.com/zyqin19/Video Hi Graph. |
| Researcher Affiliation | Academia | Zheyun Qin1, Xiankai Lu1 , Xiushan Nie2, Yilong Yin1*, Jianbing Shen3 1 School of Software, Shandong University 2 School of Computer Science and Technology, Shandong Jianzhu University 3 SKL-IOTSC, CIS, University of Macau |
| Pseudocode | No | No explicit pseudocode or algorithm blocks (e.g., labeled 'Algorithm' or 'Pseudocode') were found. |
| Open Source Code | Yes | Our algorithm implementations have been made publicly available at https://github.com/zyqin19/Video Hi Graph. |
| Open Datasets | Yes | Our training videos come from Kinetics400 (Carreira and Zisserman 2017) without using any annotation labels. |
| Dataset Splits | Yes | Quantitative comparisons for part segmentation and pose tracking on VIP and JHMDB val, respectively. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments. It mentions using 'ResNet-18' as an encoder but no information about the computational resources. |
| Software Dependencies | No | The paper mentions 'Adam solver' but does not provide specific version numbers for any software dependencies (e.g., libraries, frameworks like PyTorch, TensorFlow, or specific Python versions). |
| Experiment Setup | Yes | Implementation: We set T/2 = 9 for the distant sampling strategy of the graph-level correspondence, and node-level continuous sampling (Xu and Wang 2021) is set with a fixed frame interval of 2 from the starting frame. The hyper-parameters are empirically set to: H = 5, P =3, κ = 0.03, τp = 0.1, τa = 0.07, β = 0.5. We use the Adam solver to optimize the loss function. The learning rate is set to 1 10 4, and the weight decay is 1 10 6. |