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..
Weak-shot Keypoint Estimation via Keyness and Correspondence Transfer
Authors: Junjie Chen, Zeyu Luo, Zezheng Liu, Wenhui Jiang, Li Niu, Yuming Fang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments and analyses on large-scale benchmark MP-100 demonstrate our effectiveness. We conduct comprehensive experiments and in-depth analyses on the large-scale multi-category pose dataset MP-100 [69]. Quantitative comparison. We summarize the results of above baselines and our method on five dataset splits in Tab. 1. |
| Researcher Affiliation | Academia | 1Jiangxi University of Finance and Economics 2Shanghai Jiao Tong University |
| Pseudocode | No | The paper describes its methodology using textual descriptions and mathematical equations, but it does not contain any clearly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | We will release the code after the acceptance of paper. |
| Open Datasets | Yes | We follow the related works [69, 59, 9] to conduct our experiments on MP-100 dataset [69], which is the most prevalent benchmark dataset and covers 100 classes within 8 super-classes. |
| Dataset Splits | Yes | We follow the dataset splits in [69], i.e., all classes are split into non-overlapping train/val/test sets with the ratio of 70 : 10 : 20, and there are five random splits (S1-S5) to reduce the impact of randomness. |
| Hardware Specification | No | The paper states in its checklist that details of computer resources are provided in the appendix, however, the provided paper text does not contain an appendix with specific hardware details (e.g., GPU models, CPU types, memory amounts) used for experiments. |
| Software Dependencies | No | The paper mentions building upon a "pretrained diffusion model [55]" and using "the architecture of [44]", but it does not provide specific version numbers for any software libraries, programming languages, or other dependencies (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | Unless otherwise stated, we use 3 labeled images per base class and 30 unlabeled images per novel class by default. As for the loss balancing, we find that LBP balance well with LBK, may due to the same map representation. Besides, α = 0.1, β = 0.2, γ = 0.1 and λ = 0.5 are hyper-parameters for balancing the other losses. We compute the query as Qc l = Φc l (zt=1) RH W Dl, where z is the latent embedding mapped from image x, and we use t = 1 as suggested by [18]. We use the same configuration and denote as LNU. |