Consolidator: Mergable Adapter with Group Connections for Visual Adaptation
Authors: Tianxiang Hao, Hui Chen, Yuchen Guo, Guiguang Ding
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To verify the superiority of consolidator, we conduct extensive experiments and analysis on a series of downstream recognition tasks. Experimental results show that our consolidator can surpass full fine-tuning by 7.56 top-1 accuracy with merely 0.35% parameters per task. |
| Researcher Affiliation | Academia | Tianxiang Hao1,3, Hui Chen2,3 , Yuchen Guo3, Guiguang Ding1,3 1School of Software, Tsinghua University 2Department of Automation, Tsinghua University 3Beijing National Research Center for Information Science and Technology (BNRist) |
| Pseudocode | No | The paper describes its methods using text and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at github. |
| Open Datasets | Yes | VTAB-1k (Zhai et al., 2019) benchmark, which covers a wide range of visual domains in 19 datasets. ...Caltech101 (Fei-Fei et al., 2004), Cifar10 (Krizhevsky et al., 2009), Cifar100 (Krizhevsky et al., 2009))... (Wah et al., 2011), Oxford Flowers (Nilsback & Zisserman, 2008), Oxford Pets (Parkhi et al., 2012), Stanford Dogs (Khosla et al., 2011)), textures (DTD (Cimpoi et al., 2014)), scene classification (SUN397 (Xiao et al., 2010)) and satellite images (Euro SAT (Helber et al., 2019)). |
| Dataset Splits | Yes | Each dataset is divided into the training set (800 images), validation set (200 images), and test set (the original set). The final adaptation accuracy is reported on the test set. All models are fine-tuned on the full 1000 labeled images, i.e., train+val set. The validation set is used for tuning some hyperparameters." and "we select 10 widely-used datasets for visual recognition in various domains and use the original training, validation, and test split for experiments. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions software like 'mmdetection' and 'mmsegmentation' but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | On VTAB-1k, we follow the hyperparameters in VPT (Jia et al., 2022) for full fine-tuning, Head, Bias, and VPT, and mainly follow the hyperparameters in NOAH (Zhang et al., 2022) and SSF (Lian et al., 2022b) for adapter, Lo RA, NOAH, and consolidator. The detailed hyperparameters for each tuning method can be found in Tab. 7. On full data setting, we do a quick grid search to choose a proper set of training hyperparameters based on the performance of full fine-tuning for every well-trained visual representation. All training hyperparameters are shown in Tabs. 7 and 8. |