Instance LSeg - exploring instance level information from visual language model

• Main Author: Lin, Zixing
• Other Authors: Lin Guosheng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/171917
• Institution: Nanyang Technological University

This final year project explores the potential of using large-scale pretrained visual language models in instance-level zero-shot computer vision tasks. Specifically, we propose Instance LSeg - a novel approach to extend the zero-shot semantic segmentation method LSeg to perform language guided instance segmentation and grounding of natural language expressions in images. To evaluate our method, we used three popular referring datasets, and we observe that our method achieves highly competitive results against published generalized visual grounding baselines