The supplementary material is organized as follows:1 Implementation Details.(Sec. Except for the Object Mask Prompt and Compositional Prompt Tuning designs,7 we follow the default architecture in the original ZSseg paper. The number of part queries is set to 50.8 All the two-stage baselines are trained with AdamW optimizer with the initial learning rate of 1e-49 and weight decay of 1e-4. A poly learning rate policy with a power of 0.9is adopted.

Furthermore, the large-scale vision and language models, which play a key role in the open vocabulary setting, struggle to recognize parts as effectively as objects. To comprehensively investigate and tackle these challenges, we propose an Open-Vocabulary Part Segmentation (OV-PARTS) benchmark. OV-PARTS includes refined versions of two publicly available datasets: Pascal-Part-116 and ADE20K-Part-234.

The number of part queries is set to 50. SGD optimizer with the initial learning rate of 2e-2 and weight decay of 5e-4 is used. We sample 128 training samples for each object part class. The initial value of the learnable fusion weight is 0.5 . The total batch size is 8, and the training iterations amount to 40k.

OV -P ARTS includes refined versions of two publicly available datasets: Pascal-Part-116 and ADE20K-Part-234.

Through extensive experiments, our model demonstrated a significant improvement over the state-of-the-art models on the Pascal-Part-116, ADE20K-Part-234, and PartImageNet datasets.