Bridge the Points: Graph-based Few-shot Segment Anything Semantically

Authors: Anqi Zhang, Guangyu Gao, Jianbo Jiao, Chi Liu, Yunchao Wei

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental analysis across tasks including the standard FSS, One-shot Part Segmentation, and Cross Domain FSS validate the effectiveness and efficiency of the proposed approach, surpassing state-of-the-art generalist models with a m Io U of 58.7% on COCO-20i and 35.2% on LVIS-92i.
Researcher Affiliation Academia Anqi Zhang1, Guangyu Gao1 , Jianbo Jiao2, Chi Harold Liu1, and Yunchao Wei3 1School of Computer Science, Beijing Institute of Technology 2The MIx group, School of Computer Science, University of Birmingham 3WEI Lab, Institute of Information Science, Beijing Jiaotong University
Pseudocode Yes We mention the Mask Growth algorithm in Sec. 4.3. The Mask Growth algorithm is designed for each cluster of masks Mweak,p. The details of the algorithm are shown in Alg. 1.
Open Source Code Yes The project page of this work is: https://andyzaq.github.io/GF-SAM/. Our code and instructions are included in the supplementary material.
Open Datasets Yes Pascal-5i [22], COCO-20i [23], FSS-1000 [24], and LVIS-92i [13] are standard FSS datasets.
Dataset Splits Yes FSS-1000 [24] contains 1000 classes. The training, validation, and testing folds contain 520, 240, and 240 classes, respectively.
Hardware Specification Yes All experiments are conducted on a single NVIDIA RTX2080Ti.
Software Dependencies No The paper mentions using DINOv2 and SAM as models but does not list specific software dependencies with version numbers (e.g., Python, PyTorch, or CUDA versions).
Experiment Setup Yes The input image sizes are set to 518 518 for DINOv2 and 1024 1024 for SAM following Matcher [13]. Except for the default hyperparameters of SAM and DINOv2, our approach does not have any external hyperparameter.