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].

GenEx: Generating an Explorable World

Authors: TaiMing Lu, Tianmin Shu, Alan Yuille, Daniel Khashabi, Jieneng Chen

ICLR 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 EXPERIMENTS 5.1 DATASET CONSTRUCTION 5.2 EVALUATION ON GENERATION QUALITY 5.3 EVALUATION ON IMAGINATIVE EXPLORATION QUALITY 5.4 RESULTS ON EMBODIED QA We adopt FVD (Unterthiner et al., 2019), SSIM (Wang et al., 2004), LPIPS (Zhang et al., 2018), and PSNR (Hor e & Ziou, 2010) to evaluate video generation quality
Researcher Affiliation Academia Taiming Lu, Tianmin Shu, Alan Yuille, Daniel Khashabi, Jieneng Chen Johns Hopkins University EMAIL
Pseudocode No The paper describes the methodology in prose and diagrams (e.g., Figure 3) but does not include explicit pseudocode or algorithm blocks.
Open Source Code Yes Code https://github.com/Beckschen/Gen Ex
Open Datasets Yes We collect an additional test set of panoramic images from Google Maps Street View (header Street in Table 2) and Behavior Vision Suite (Ge et al., 2024) (header Indoor in Table 2)
Dataset Splits No The paper mentions training on 'Gen Ex-DB' and evaluating on 'Gen Ex-EQA' and other test sets like Google Maps Street View and Behavior Vision Suite, but it does not specify explicit training/validation/test dataset splits with percentages, sample counts, or citations to predefined splits.
Hardware Specification Yes Total GPU Usage 384 A100 hours GPU Configuration 2 A100 per batch, Model Parallelism
Software Dependencies No The paper mentions software like 'Unity', 'Blender', 'Unreal Engine', 'GPT-4o', but does not provide specific version numbers for any of them.
Experiment Setup Yes Table 5: Gen Ex-Diffuser Training configuration. learning rate 1e-5 lr scheduler Cosine output height 576 output width 1024 mixed precision fp16 training frame 25 lr warmup steps 500