Uncertainty-Aware Learning for Zero-Shot Semantic Segmentation

By Ping Hu, Stan Sclaroff, Kate Saenko. 2020

79 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Language-driven Semantic Segmentation (2022, background): “Hu et al. (2020) propose to use uncertainty-aware learning to better handle noisy labels of seen classes, while Li et al. (2020b) introduce a structured learning approach to better exploit the relations between seen and unseen categories.”
Open-Vocabulary Image Segmentation (2021, methodology): “An alternative approach towards open-vocabulary image segmentation is zero-shot image segmentation [4, 6, 19, 28, 47].”
Extract Free Dense Labels from CLIP (2021, background): “Other studies like [26], [34], and [20], improve the generative ZS3Net in terms of uncertainty, structural consistency, and context, respectively, while STRICT [41] boosts the SPNet through self-training.”
A Simple Baseline for Open-Vocabulary Semantic Segmentation with Pre-trained Vision-Language Model (2021, background): “In [24], the uncertainty-aware losses are proposed to eliminate noisy samples.”
Scaling Open-Vocabulary Image Segmentation with Image-Level Labels (2021, methodology): “Zero-shot semantic segmentation methods [6,49,20,30,4] aim to segment images with unseen visual concepts using language embeddings.”
Open-Vocabulary Instance Segmentation via Robust Cross-Modal Pseudo-Labeling (2021, methodology): “Naively, we can train the student model to mimic the exact pseudo masks at each pixel as: where BCE is the binary cross-entropy loss for pixel logit predictions, M xyo is the pseudo masks at pixel ( x, y ) and g xy Mask is the student mask predictions at the pixel.”
Conterfactual Generative Zero-Shot Semantic Segmentation (2021, background): “In zero-shot semantic segmentation, this bias represents intuitively by the imbalance of the accuracy between seen classes and unseen classes [18].”
CLIP on Wheels: Zero-Shot Object Navigation as Object Localization and Exploration (2022, methodology): “Zero-shot techniques have become increasingly popular in image classification [53, 33, 51], object detection [4, 54, 19, 38, 77, 42, 26], and semantic segmentation [6, 35, 32, 3, 14, 76].”
DenseCLIP: Extract Free Dense Labels from CLIP (2021, background): “Other studies like [26], [34], and [20], improve the generative ZS3Net in terms of uncertainty, structural consistency, and context, respectively, while STRICT [41] boosts the SPNet through self-training.”
Image Segmentation Using Text and Image Prompts (2021, unknown): “More recent approaches use a joint space for image features and class prototypes [28], employ a probabilistic formulation to account for uncertainty [29] or model the detection of unseen objects explicitly [30].”