Semi-supervised Keypoint Localization
Authors: Olga Moskvyak, Frederic Maire, Feras Dayoub, Mahsa Baktashmotlagh
ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our semi-supervised approach significantly outperforms previous methods on several benchmarks for human and animal body landmark localization. (Abstract) and 4 EXPERIMENTS (Section Heading) |
| Researcher Affiliation | Academia | Olga Moskvyak, Frederic Maire, Feras Dayoub School of Electrical Engineering and Robotics Queensland University of Technology, Australia {olga.moskvyak,f.maire,feras.dayoub}@qut.edu.edu Mahsa Baktashmotlagh School of Information Technology and Electrical Engineering The University of Queensland, Australia m.baktashmotlagh@uq.edu.au |
| Pseudocode | No | The paper describes the method and components (e.g., in Section 3 and Figure 1, Figure 2) but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or link for the release of its own source code. |
| Open Datasets | Yes | MPII Human Pose dataset (Andriluka et al., 2014) is a collection of images... LSP (Leeds Sports Pose) (Johnson & Everingham, 2010; 2011) dataset is a collection of annotated images... CUB-200-2011 (Welinder et al., 2010) is a dataset of 200 fine-grained classes of bird species. ATRW (Li et al., 2019) is a dataset of Amur tigers images... |
| Dataset Splits | Yes | Training set for each dataset is split into labeled and unlabeled subsets by randomly picking 5%, 10%, 20% or 50% of the training examples and discarding the labels for the rest of the data. The procedure is repeated three times so all experiments are run three times to obtain the mean and standard deviation of the results. Validation and test sets are fixed for all experiments. Validation set is used to tune hyperparameters and test set is used to report the final results. (Section 4.1) and Our validation and test sets consist of 3,311 and 2,958 images respectively. (for MPII, Section 4.1) |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as exact GPU or CPU models. |
| Software Dependencies | No | The paper mentions tools and architectures like HRNet-32 and Adam optimizer, but it does not specify software dependencies with version numbers (e.g., Python version, library versions like PyTorch 1.9). |
| Experiment Setup | Yes | Images are resized to the input size 256 256 and heatmaps are predicted at size 64 64. ... We adopt Adam (Kingma & Ba, 2015) optimizer with learning rate 10^-4 for all experiments. ... The weights of loss components were determined experimentally (λ1, λ2, λ3, λ4) = (10^3, 0.5, 10^2, 10^2). |