Tool
BeetleFlow: An Integrative Deep Learning Pipeline for Beetle Image Processing
Pipeline for processing large-scale tray images of pinned beetles: detect specimens, optionally sort and crop with metadata matching, and produce fine-grained morphological segmentation (5-class or 9-class). It uses an open-vocabulary detector (Grounding DINO), a vision-language model (LLaVA-NeXT) for verification, and fine-tuned Mask2Former for transformer-based part segmentation and downstream morphological analysis.
Liu, Fangxun, Rayeed, S M, Stevens, Samuel, East, Alyson, Chiang, Cheng Hsuan, Lee, Colin, Yi, Daniel, Yang, Junke, Naik, Tejas, Wang, Ziyi, Kilrain, Connor, Buckwalter, Elijah H., Hou, Jiacheng, Bueno, Saul Ibaven, Wang, Shuheng, Ma, Xinyue, Liu, Yifan, Tao, Zhiyuan, Zhang, Ziheng, Sokol, Eric, Belitz, Michael, Record, Sydne, Stewart, Charles V., Chao, Wei-Lun
mit-licenseView resource