Frame-Level Label Refinement for Skeleton-Based Weakly-Supervised Action Recognition
Authors: Qing Yu, Kent Fujiwara
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We extensively evaluate the effectiveness of our learning approach on skeleton-based action recognition benchmarks. The state-of-the-art experimental results demonstrate that the proposed method can recognize and localize action segments of the skeleton data. We evaluate our method on a skeleton-based AR benchmark, BABEL (Punnakkal et al. 2021). In many settings, our method outperforms existing methods by a large margin. |
| Researcher Affiliation | Collaboration | Qing Yu,1 Kent Fujiwara2 1The University of Tokyo, Japan 2LINE Corporation, Japan |
| Pseudocode | Yes | Algorithm 1: Joint Optimization |
| Open Source Code | No | The paper does not provide any explicit statement or link regarding the release of its own source code. |
| Open Datasets | Yes | We verify the effectiveness of our approach on the benchmark dataset of 3D human motion, BABEL (Punnakkal et al. 2021). |
| Dataset Splits | No | Table 2 provides '#Training Sequences' and '#Test Sequences' but does not explicitly state a 'validation' dataset split with numbers or percentages. |
| Hardware Specification | Yes | We use the Py Torch library (Paszke et al. 2019) to implement the proposed framework on a single NVIDIA A100 GPU. |
| Software Dependencies | No | The paper mentions 'Py Torch library (Paszke et al. 2019)' but does not provide specific version numbers for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | The Adam optimizer with the learning rate of 0.0001 is applied to optimize the network at the mini-batch level with batch-size 8 for 100 epochs. The hyper-parameters adopted to construct the classifier are empirically set as follows: δ = 5.0, τ = 2.0. The α used for selecting clean samples in Eq. (13) is set as 0.2. The loss weights λpseudo and λKL are set to 1.0 and 0.5, respectively. We begin updating the frame-level pseudo labels from the 10th epoch... |