Multi-Granularity Video Object Segmentation

Authors: Sangbeom Lim, Seongchan Kim, Seungjun An, Seokju Cho, Paul Hongsuck Seo, Seungryong Kim

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Additionally, we evaluate existing video segmentation methods and MMPM on MUG-VOS test dataset. MMPM shows best performance quantitatively and qualitatively on MUG-VOS dataset. We conducted ablation studies on the MMPM model using the MUG-VOS test dataset.
Researcher Affiliation Collaboration Sangbeom Lim1*, Seongchan Kim1*, Seungjun An2*, Seokju Cho3, Paul Hongsuck Seo1 , Seungryong Kim3 1Korea University 2Samsung Electronics 3KAIST
Pseudocode No The paper describes methods using equations and diagrams (Figure 2 and Figure 3) but does not contain any explicitly labeled pseudocode or algorithm blocks. For instance, the 'Data collection pipeline' section provides mathematical formulas (Equations 1-6) to describe the process.
Open Source Code Yes Code https://cvlab-kaist.github.io/MUG-VOS
Open Datasets Yes We propose a large-scale, densely annotated multi-granularity video object segmentation (MUG-VOS) dataset that includes various types and granularities of mask annotations. Code https://cvlab-kaist.github.io/MUG-VOS
Dataset Splits Yes MUG-VOS Train 0.714 47M 4.7M 66.3 77,9940 MUG-VOS Test 0.663 59K 887 29.6 1,999 (from Table 1). All models are trained on the MUG-VOS train set.
Hardware Specification No This research was supported by... the National Supercomputing Center with supercomputing resources including technical support (KSC-2023-CRE-0416). This statement mentions a supercomputing resource but does not specify details such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper does not provide specific versions for any software dependencies, programming languages, or libraries used in the implementation or experimentation.
Experiment Setup Yes We conducted ablation studies on the MMPM model using the MUG-VOS test dataset. These studies explore the effects of different memory filtering methods, the number of memory values, memory update intervals, and the characteristics of each memory module. Table 3a shows the performance of the MMPM model with different memory filtering methods. Table 3b shows the performance comparison between different memory usage types. Table 4a shows the performance variation based on the interval period, denoted as r... Table 4b shows the performance variation when implementing different number of memory values in the temporal memory, denoted as N.