BarLeRIa: An Efficient Tuning Framework for Referring Image Segmentation

Authors: Yaoming Wang, Jin Li, XIAOPENG ZHANG, Bowen Shi, Chenglin Li, Wenrui Dai, Hongkai Xiong, Qi Tian

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments conducted on RIS benchmarks demonstrate the superiority of Bar Le RIa over prior PET methods with a significant margin, i.e., achieving an average improvement of 5.6%.
Researcher Affiliation Collaboration 1Shanghai Jiao Tong University 2Huawei Cloud
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code is available at https://github.com/Nastrond Ad/Bar Le RIa.
Open Datasets Yes Datasets. We employ three challenging referring image segmentation benchmarks in our experiments: Ref COCO (Kazemzadeh et al., 2014), Ref COCO+ (Kazemzadeh et al., 2014), and G-Ref (Yu et al., 2016). Please refer to the appendix A.1 for details.
Dataset Splits Yes Ref COCO (Kazemzadeh et al., 2014)... The dataset is divided into four subsets: 120,624 samples for training, 10,834 for validation, 5,657 for test A, and 5,095 for test B.
Hardware Specification Yes We train the model using 2 Tesla V100 GPUs with a batch size of 32. For Vi T-L/14, we train the model using 8 Tesla V100 GPUs with a batch size of 64 and an initial learning rate of 0.0002.
Software Dependencies No The paper mentions using the Adam optimizer but does not specify software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions).
Experiment Setup Yes We train the whole network in an end-to-end manner for 50 epochs using the Adam optimizer with a learning rate of 0.0001. A learning rate decay is employed at the 35th epoch with a decay factor of 0.1.