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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
VideoTitans: Scalable Video Prediction with Integrated Short- and Long-term Memory
Authors: Young-Jae Park, Minseok Seo, Hae-Gon Jeon
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on Moving-MNIST, Human3.6M, Traffic BJ and Weather Bench benchmarks show that Video Titans consistently reduces computation (FLOPs) and achieves competitive visual fidelity compared to state-of-the-art recurrent, convolutional, and efficient-transformer methods. Comprehensive ablations confirm that each proposed component contributes significantly. |
| Researcher Affiliation | Academia | Young-Jae Park Department of AI Convergence GIST EMAIL Minseok Seo School of Electrical Engineering KAIST EMAIL Hae-Gon Jeon Department of Artificial Intelligence Yonsei University EMAIL |
| Pseudocode | No | The paper describes the methodology and architecture of Video Titans in text and mathematical formulations (Equations 1-8) but does not include a distinct 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | To facilitate further research and ensure full reproducibility, we will publicly release our source code, trained checkpoints, and demonstration videos. Additionally, in the NeurIPS Paper Checklist, section 5, the authors state: 'Answer: [Yes] Justification: The implementation details and code are provided to ensure reproducibility.' |
| Open Datasets | Yes | We evaluate Video Titans on four widely-used datasets for future frame prediction, summarized in Table 2: Moving MNIST [53], Human3.6M [26], Traffic BJ [74], and Weather Bench [45]. |
| Dataset Splits | Yes | Dataset Ntrain Ntest (C, H, W) T T Moving MNIST [53] 10,000 10,000 (1, 64, 64) 10 10 Traffic BJ [74] 20,461 500 (2, 32, 32) 4 4 Human3.6 [26] 73,404 8,582 (3, 128, 128) 4 4 Weatherbench [45] 2,167 706 (1/2, 32, 64) 12 12 |
| Hardware Specification | Yes | All experiments are implemented using Py Torch and conducted on 8 NVIDIA A100 GPUs. |
| Software Dependencies | No | The paper mentions implementing experiments using PyTorch and utilizing the Adam optimizer and Reduce LROn Plateau scheduler, but does not provide specific version numbers for PyTorch or other libraries. |
| Experiment Setup | Yes | Following [58], we optimize Video Titans using the Adam optimizer and train with the Mean Squared Error (MSE) loss. We set the batch size to 8 for all experiments. The learning rate is adaptively adjusted using the Reduce LROn Plateau scheduler with patience of 10 epochs. Initial learning rates are selected from the set {10 2, 5 10 3, 10 3, 5 10 4, 10 4}, and the best-performing value is used for each dataset. The total number of training epochs varies depending on the dataset complexity and size. Table 4: Detailed hyperparameter configuration used for Video Titans training. |