MiddleCLIP: Unleashing the potential of the middle layers in CLIP for training-free open-vocabulary semantic segmentation

Xiao Jin, Dai-Wei Chen, Wei Shi, Zhuang Shao. MiddleCLIP: Unleashing the potential of the middle layers in CLIP for training-free open-vocabulary semantic segmentation. Neurocomputing, 698:134289, 2026. [doi]

@article{JinCSS26,
  title = {MiddleCLIP: Unleashing the potential of the middle layers in CLIP for training-free open-vocabulary semantic segmentation},
  author = {Xiao Jin and Dai-Wei Chen and Wei Shi and Zhuang Shao},
  year = {2026},
  doi = {10.1016/j.neucom.2026.134289},
  url = {https://doi.org/10.1016/j.neucom.2026.134289},
  researchr = {https://researchr.org/publication/JinCSS26},
  cites = {0},
  citedby = {0},
  journal = {Neurocomputing},
  volume = {698},
  pages = {134289},
}