Unsupervised Learning of Compositional Scene Representations from Multiple Unspecified Viewpoints

Authors: Jinyang Yuan, Bin Li, Xiangyang Xue8971-8979

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on several specifically designed synthetic datasets have shown that the proposed method is able to effectively learn from multiple unspecified viewpoints.
Researcher Affiliation Academia Shanghai Key Laboratory of Intelligent Information Processing, School of Computer Science, Fudan University {yuanjinyang, libin, xyxue}@fudan.edu.cn
Pseudocode Yes Algorithm 1: Inference of Latent Variables
Open Source Code Yes 1Code is available at https://git.io/JDnne.
Open Datasets Yes Datasets: The experiments are performed on four multiviewpoint variants (referred to as CLEVR-M1 to CLEVRM4) of the commonly used CLEVR dataset that differ in the ranges to sample viewpoints and in the attributes of objects. Further details are described in the Supplementary Material. ...constructed based on the d Sprites (Matthey et al. 2017), Abstract Scene (Zitnick and Parikh 2013), and CLEVR (Johnson et al. 2017) datasets, in a way similar to the Multi-Objects Datasets (Kabra et al. 2019) but provides extra annotations (for evaluation only) of complete shapes of objects.
Dataset Splits No The paper states that 'All the methods are trained and tested with M = 4 and K = 7' and 'All the methods are trained and tested with K =6, K =5, and K =7 on the d Sprites, Abstract, and CLEVR datasets, respectively.' However, it does not provide specific percentages or counts for training, validation, and test splits, nor does it refer to predefined splits with citations that contain this information.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies, such as library or solver names with version numbers, needed to replicate the experiment.
Experiment Setup No While the paper mentions hyperparameters such as λ and α, it does not provide concrete values for these or other common experimental setup details like learning rate, batch size, number of epochs, or optimizer settings in the main text.