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..
FLOWING: Implicit Neural Flows for Structure-Preserving Morphing
Authors: Arthur Bizzi, Matias Grynberg Portnoy, Vitor Pereira Matias, Daniel Perazzo, João Paulo Silva do Monte Lima, Luiz Velho, Nuno Gonçalves, João Pereira, Guilherme Schardong, Tiago Novello
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate FLOWING on face and image morphing, as well as on 3DGS morphing, using four diverse datasets. These experiments demonstrate the effectiveness of our approach in both 2D and 3D settings. Additionally, we provide ablation studies in Appendix B to validate our architectural choices and regularization strategies. |
| Researcher Affiliation | Academia | 1EPFL 2University of Buenos Aires 3University of São Paulo 4IMPA 5CSML-IIT 6Universidade Federal Rural de Pernambuco 7ISR-UC 8INCM 9University of Georgia |
| Pseudocode | No | The paper describes methods and mathematical formulations but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks with structured, code-like steps. Appendix A details forward differentiation but not in pseudocode format. |
| Open Source Code | Yes | Code and pretrained models are available in https://schardong.github.io/flowing. |
| Open Datasets | Yes | For face images, we use the FRLL dataset [10], which contains 102 identities with 5 images captured at fixed angles and two expressions (neutral and smiling). The second dataset is a subset of Mega Depth [25], which provides multi-view scene landmarks. Third, we use eight subjects from Ne RSemble [22], a multi-view collection of human heads. Finally, we include in-the-wild face images from FFHQ [18] for qualitative evaluation. |
| Dataset Splits | No | The paper mentions using specific datasets (FRLL, Mega Depth, Ne RSemble, FFHQ) and describes the total number of items or how pairs were constructed, but it does not provide explicit training, validation, or test splits (e.g., percentages, sample counts, or references to standard splits) for reproducing the experimental partitioning of the data. |
| Hardware Specification | Yes | Table 3 summarizes training and morphing times obtained on an RTX 4090 GPU for both warp training and morphing (warping inference + blending). |
| Software Dependencies | No | The paper states: "FLOWING is implemented in Py Torch [33]". While it cites PyTorch and implies a version by publication year (2019), it does not provide a specific version number for PyTorch or any other software dependency, such as "PyTorch 1.9" or "Python 3.8". |
| Experiment Setup | Yes | FLOWING is implemented in Py Torch [33] and trained with the Adam optimizer [21]. At each training step, we sample 20,000 points from the spatial domain [ 1, 1]2 for the selected values of t. The initial LRs are set to 0.001 for NODE, 0.002 for NCF, and 0.0001 for ifmorph. Early stopping is employed with a patience of 500 epochs for NCF and NODE, and 1,000 for ifmorph. If the loss plateaus for 100 epochs, the LR is reduced, and training terminates once the patience threshold is reached. |