Res-Tuning: A Flexible and Efficient Tuning Paradigm via Unbinding Tuner from Backbone

Authors: Zeyinzi Jiang, Chaojie Mao, Ziyuan Huang, Ao Ma, Yiliang Lv, Yujun Shen, Deli Zhao, Jingren Zhou

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on both discriminative and generative tasks demonstrate the superiority of our method over existing alternatives from the perspectives of efficacy and efficiency.
Researcher Affiliation Collaboration 1Alibaba Group 2National University of Singapore 3Ant Group
Pseudocode No The paper includes equations and architectural diagrams but does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Project page: https://res-tuning.github.io/.
Open Datasets Yes For most experiments, we adopt Vi T-B/16 [13] pre-trained on Image Net-21K [11] as the backbone model, following VPT [28]. ... We evaluate the text-to-image generation performance on COCO2017 dataset [43]. ... Table 8: Datasets used for generative tasks. ... Table 9: Datasets used for discriminative tasks.
Dataset Splits No The paper mentions using 'validation set' for evaluation (e.g., 'combine the validation set of 19 tasks in VTAB-1K' or 'sample 10k captions from the validation set'), but it does not consistently provide specific train/validation/test dataset splits with percentages or counts for all experiments.
Hardware Specification Yes Device A100 1 (for discriminative tasks), Device A100 8 (for generative tasks)
Software Dependencies Yes Library Diffusers 2
Experiment Setup Yes Table 10: Hyperparameter selection for discriminative tasks. ... Table 11: Hyperparameter selection for generative tasks.