Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Object-Centric Representation Learning with Generative Spatial-Temporal Factorization
Authors: Nanbo Li, Muhammad Ahmed Raza, Wenbin Hu, Zhaole Sun, Robert Fisher
NeurIPS 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We used two simulated multi-view-dynamic-scene synthetic datasets, namely DRoom and MJCArm, and a real-world dataset, namely Cube Land (see Appendix C.3 for details), in this work. We conducted quantitative analysis on DRoom and show qualitative results on the other two datasets. |
| Researcher Affiliation | Academia | Li Nanbo School of Informatics University of Edinburgh EMAIL Muhammad Ahmed Raza School of Informatics University of Edinburgh EMAIL Hu Wenbin School of Informatics University of Edinburgh EMAIL Zhaole Sun School of Informatics University of Edinburgh EMAIL Robert B. Fisher School of Informatics University of Edinburgh EMAIL |
| Pseudocode | Yes | Algorithm 1: Dy MON Training Algorithm |
| Open Source Code | No | The paper does not provide an explicit statement or link for open-source code availability. |
| Open Datasets | Yes | We used two simulated multi-view-dynamic-scene synthetic datasets, namely DRoom and MJCArm, and a real-world dataset, namely Cube Land (see Appendix C.3 for details), in this work. |
| Dataset Splits | Yes | The DRoom dataset consists of five subsets (including both training and testing sets): one subset (denoted as DR0-|fz|) with zero object motion (multi-view-static-scene data), one subset (denoted as DR0-|fv|) with zero camera motion (single-view-dynamic-scene data), and three multi-view-dynamic-scene subsets of increasing speed difference levels from 1 to 3 (denoted as DR-Lvl.1 3). Each of the five subsets consists of around 200 training sequences (40 frames of RGB images per sequence) and 20 testing sequences (40 frames from 12 different views, i.e. 57.6k images). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | Hyperparameters |Q|, (βF CSO, βSCF O), ( t, τ) ; // t > τ > 2, |Q| = sizeof(Q) (Algorithm 1 Input) [...] All models were trained with 3 different random seeds for quantitative comparisons. Refer to our supplementary material for full details on experimental setups, and ablation studies and more qualitative results. |