Semantic Prompt Learning for Weakly-Supervised Semantic Segmentation

TL;DR

Weakly-Supervised Semantic Segmentation typically relies on CAM-based localization with coarse pseudo masks. SemPLeS introduces semantic prompt learning that auto-discovers class-associated background semantics from the CLIP latent space using Contrastive Prompt Learning and refines segmentations with Prompt-guided Semantic Refinement, yielding more accurate CAMs and pseudo masks. The approach uses a Segment-Label Matching objective alongside learnable prompts, and a refinement loss to suppress background activations, achieving competitive results on PASCAL VOC 2012 and MS COCO 2014 and demonstrating compatibility with other WSSS methods. This work advances vision-language grounding for segmentation by eliminating manual prompting and enhancing semantic alignment between object regions and labels, with practical impact for label-efficient segmentation pipelines.

Abstract

Weakly-Supervised Semantic Segmentation (WSSS) aims to train segmentation models using image data with only image-level supervision. Since precise pixel-level annotations are not accessible, existing methods typically focus on producing pseudo masks for training segmentation models by refining CAM-like heatmaps. However, the produced heatmaps may capture only the discriminative image regions of object categories or the associated co-occurring backgrounds. To address the issues, we propose a Semantic Prompt Learning for WSSS (SemPLeS) framework, which learns to effectively prompt the CLIP latent space to enhance the semantic alignment between the segmented regions and the target object categories. More specifically, we propose Contrastive Prompt Learning and Prompt-guided Semantic Refinement to learn the prompts that adequately describe and suppress the co-occurring backgrounds associated with each object category. In this way, SemPLeS can perform better semantic alignment between object regions and class labels, resulting in desired pseudo masks for training segmentation models. The proposed SemPLeS framework achieves competitive performance on standard WSSS benchmarks, PASCAL VOC 2012 and MS COCO 2014, and shows compatibility with other WSSS methods. Code: https://github.com/NVlabs/SemPLeS.

Figures (6)

• Figure 1: An overview of our proposed SemPLeS framework. We first introduce (a) Segment-Label Matching, which leverages image-text contrastive learning to train the mask generator $S$ and produce initial object masks $M$. Such derived masks are still coarse and may falsely include co-occurring backgrounds. To achieve class-associated mask refinement and produce the refined mask $M'$, we propose (b) Contrastive Prompt Learning to automatically learn prompts $p_k$ embedded with semantic knowledge from the CLIP latent space, followed by (c) Prompt-guided Semantic Refinement to suppress co-occurring backgrounds associated with each category $k$.
• Figure 2: Qualitative results of CAMs. "GT" denotes the ground truth masks. We see that our proposed SemPLeS framework produces precise CAMs better aligned with the ground truth masks.
• Figure 3: Qualitative results of segmentation maps. "GT" denotes the ground truth masks.
• Figure 4: Visualization of the manually-defined background prompts xie2022clims and our learned prompts.
• Figure 5: Qualitative ablation studies of loss functions. With both $L_{prompt}$ and $L_{refine}$ applied, the resulting CAMs are better aligned with the ground truth masks.