Open-Vocabulary Transparent Object Segmentation
AI/MLOngoing research on open-vocabulary semantic segmentation for transparent objects, using parameter-efficient fine-tuning.
- Year
- 2026
- Role
- Research, in progress
- Stack
- PyTorch, PEFT, LoRA, Computer Vision
Status
Ongoing research project on open-vocabulary semantic segmentation for transparent objects. Complements DIWA-Net work in computer vision and PEFT/LoRA.
Focus
Transparent and reflective surfaces break many segmentation pipelines that assume opaque appearance. The goal is open-vocabulary segmentation that still holds up when the object of interest is glass, plastic, or similarly hard to see — without full fine-tuning of large vision backbones.