Reproduce “MM-CamObj: A Comprehensive Multimodal Dataset for Camouflaged Object Scenarios” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jiacheng Ruan, Wenzhen Yuan, Zehao Lin, Ning Liao, Zhiyu Li, Feiyu Xiong, Ting Liu, Yuzhuo Fu

CamObj-LLaVA adapts LLaVA-v1.5 to camouflage scenes using MM-CamObj, which provides image descriptions for alignment and question-answer conversations for instruction tuning. The authors use LoRA in both training stages and order examples by combined instruction and response length. On CamObj-Bench, the model reaches 56.7% mIoU for box localization, versus GPT-4o’s 32.0%, but gains do not extend to every task or multi-image masks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.