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..
Vocabulary-Guided Gait Recognition
Authors: Panjian Huang, Saihui Hou, Chunshui Cao, Xu Liu, Yongzhen Huang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on CASIA-B, CCPG, SUSTech1K, Gait3D and GREW reveal the potential value and research directions of vocabulary information from VLMs in the gait field. Section 4: Experiments. Table 1, 2, 3, 4 provide evaluation metrics (e.g., Rank-1, Rank-5) and comparisons with other methods. Section 4.3 is titled 'Ablation Study'. |
| Researcher Affiliation | Collaboration | 1 School of Artificial Intelligence, Beijing Normal University; 2 WATRIX.AI. Beijing Normal University is an academic institution, and WATRIX.AI is a company, indicating a collaboration. |
| Pseudocode | No | The paper describes methods using mathematical formulations like O = P(E(X)) (1) and F = G(O) (2), and also details processes such as V f = Re LU(LN(Linear(Vf))) (4). However, it does not include any clearly labeled pseudocode blocks or algorithms in a structured, code-like format. |
| Open Source Code | No | All datasets are public data, and the code will be open-access if accepted. |
| Open Datasets | Yes | Extensive experiments on CASIA-B, CCPG, SUSTech1K, Gait3D and GREW reveal the potential value and research directions of vocabulary information from VLMs in the gait field. Appendix A.1 Databases: Gait databases are commonly categorized into two groups: Constrained and In-the-wild scenarios. As shown in Table 7, CASIA-B[63], CCPG [57] generally include fewer individuals but provide explicit condition types. In-the-wild databases SUSTech1K [38], Gait3D [16] and GREW [64] contain a larger number of identities and more challenging scenarios (e.g., occlusions). |
| Dataset Splits | Yes | Appendix A.1 Databases, Table 7: Id. and Seq. denote the number of identities and sequences. For CASIA-B: Train Id. 74, Seq. 8,140; Test Id. 50, Seq. 5,500. For CCPG: Train Id. 100, Seq. 8,187; Test Id. 100, Seq. 8,095. For SUSTech1K: Train Id. 200, Seq. 5988; Test Id. 850, Seq. 19,228. For Gait3D: Train Id. 3,000, Seq. 18,940; Test Id. 1,000, Seq. 6,369. For GREW: Train Id. 20,000, Seq. 102,887; Test Id. 6,000, Seq. 24,000. |
| Hardware Specification | No | Existing methods in this field generally do not report such information, and no comparison was conducted. (From NeurIPS Paper Checklist, Question 8 response). |
| Software Dependencies | No | Appendix A.2, 'Optimization': We employ SGD with an initial learning rate of 0.1. This mentions an optimizer but does not provide specific version numbers for software libraries or programming languages used for implementation. |
| Experiment Setup | Yes | Appendix A.2 Implementation Details, Optimization: We employ SGD with an initial learning rate of 0.1, which is reduced by 0.1 at specific iteration milestones where CASIA-B, CCPG, SUSTech1K, Gait3D and GREW are [20K, 40K, 50K], [20K, 40K, 50K], [20K, 30K, 40K], [20K, 40K, 50K] and [80K, 120K, 150K], respectively. The total training iterations of CASIA-B, CCPG, SUSTech1K, Gait3D and GREW are 60K, 60K, 50K, 60K, 180K, respectively. Inputs. The silhouettes on all databases are transformed into 64 44, and each sequence consists of 30 consecutive frames. |