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..

VT-FSL: Bridging Vision and Text with LLMs for Few-Shot Learning

Authors: Wenhao Li, Qiangchang Wang, Xianjing Meng, Zhibin Wu, Yilong Yin

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted in three distinct few-shot learning scenarios.
Researcher Affiliation Academia 1School of Software, Shandong University 2Shenzhen Loop Area Institute 3School of Computing and Artificial Intelligence, Shandong University of Finance and Economics
Pseudocode Yes A.1 Training Algorithm Algorithm 1: Training algorithm of the proposed VT-FSL.
Open Source Code Yes Code is available at https://github.com/peacelwh/VT-FSL.
Open Datasets Yes Datasets. Extensive experiments are conducted in three distinct few-shot learning scenarios. (1) Four datasets in standard FSL: mini Image Net [12], tiered Tiered Net [60], CIFAR-FS [67], and FC100 [19]. (2)Three datasets in fine-grained FSL: CUB [68], Cars [69], and Dogs [70]. (3)Three datasets in cross-domain FSL: CUB, Places [71], and Plantae [72].
Dataset Splits Yes Table 8: The splits of categories and the number of categories/images in each few-shot dataset. ... mini Image Net [12] 64 16 20 12,000 60,000
Hardware Specification Yes All experiments are performed with an NVIDIA RTX 6000 Ada.
Software Dependencies No The paper mentions specific models (Visformer Tiny, CLIP, Qwen2.5-VL-32B, Janus-Pro) and an optimizer (Adam W), but does not list specific version numbers for general software dependencies such as Python, PyTorch, or TensorFlow.
Experiment Setup Yes The Adam W optimizer [73] is used with a learning rate of 5e-4 and a cosine scheduler. Pre-training runs for 300 epochs in tiered Image Net and 800 epochs in other datasets with a batch size of 512, followed by meta-tuning for 100 epochs via an episodic training strategy. The hyperparameter τ is set as 0.2 according to validation accuracy.