Language-driven Semantic Segmentation

Authors: Boyi Li, Kilian Q Weinberger, Serge Belongie, Vladlen Koltun, Rene Ranftl

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct quantitative evaluation on a variety of zeroand few-shot semantic segmentation tasks.
Researcher Affiliation Collaboration Boyi Li Cornell University, Cornell Tech Kilian Q. Weinberger Cornell University Serge Belongie University of Copenhagen Vladlen Koltun Apple René Ranftl Intel Labs
Pseudocode No The paper describes the method and architecture using natural language and diagrams, but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes Code and demo are available at https://github.com/isl-org/lang-seg.
Open Datasets Yes We evaluate on three widely-used few-shot semantic segmentation benchmarks: PASCAL-5i (Everingham et al., 2015), COCO-20i (Lin et al., 2014), and FSS-1000 (Li et al., 2020c).
Dataset Splits Yes Following the standard protocol, we split the 1000 classes into training, validation, and test classes, with 520, 240, and 240 classes, respectively.
Hardware Specification Yes We train with a batch size of 6 on six Quadro RTX 6000.
Software Dependencies No The paper mentions software like PyTorch and refers to various models and frameworks (e.g., CLIP, ViT, ResNet, DPT), but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes We use SGD with momentum 0.9 and a polynomial learning rate scheduler with decay rate 0.9. We train with a batch size of 6 on six Quadro RTX 6000.