Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Promoting SAM for Camouflaged Object Detection via Selective Key Point-based Guidance

Guoying Liang, Su Yang

TL;DR

This paper reframes Camouflaged Object Detection (COD) as a prompt-guided segmentation task by leveraging the Segment Anything Model (SAM) with selective point-based prompts. It introduces PPT-net to predict per-point probabilities on a grid and a Key Point Selection (KPS) algorithm to assemble positive and negative point triples that guide SAM without altering its weights. The approach achieves competitive or state-of-the-art results on COD10K, NC4K, and CAMO, and ablation studies validate the impact of promotions, grid granularity, and the two-pass leveraging pipeline. The work offers an efficient, generalizable pathway to apply large, pre-trained segmentation models to COD, reducing task-specific design while maintaining strong performance and interpretability through informative promotions.

Abstract

Big model has emerged as a new research paradigm that can be applied to various down-stream tasks with only minor effort for domain adaption. Correspondingly, this study tackles Camouflaged Object Detection (COD) leveraging the Segment Anything Model (SAM). The previous studies declared that SAM is not workable for COD but this study reveals that SAM works if promoted properly, for which we devise a new framework to render point promotions: First, we develop the Promotion Point Targeting Network (PPT-net) to leverage multi-scale features in predicting the probabilities of camouflaged objects' presences at given candidate points over the image. Then, we develop a key point selection (KPS) algorithm to deploy both positive and negative point promotions contrastively to SAM to guide the segmentation. It is the first work to facilitate big model for COD and achieves plausible results experimentally over the existing methods on 3 data sets under 6 metrics. This study demonstrates an off-the-shelf methodology for COD by leveraging SAM, which gains advantage over designing professional models from scratch, not only in performance, but also in turning the problem to a less challenging task, that is, seeking informative but not exactly precise promotions.

Promoting SAM for Camouflaged Object Detection via Selective Key Point-based Guidance

TL;DR

This paper reframes Camouflaged Object Detection (COD) as a prompt-guided segmentation task by leveraging the Segment Anything Model (SAM) with selective point-based prompts. It introduces PPT-net to predict per-point probabilities on a grid and a Key Point Selection (KPS) algorithm to assemble positive and negative point triples that guide SAM without altering its weights. The approach achieves competitive or state-of-the-art results on COD10K, NC4K, and CAMO, and ablation studies validate the impact of promotions, grid granularity, and the two-pass leveraging pipeline. The work offers an efficient, generalizable pathway to apply large, pre-trained segmentation models to COD, reducing task-specific design while maintaining strong performance and interpretability through informative promotions.

Abstract

Big model has emerged as a new research paradigm that can be applied to various down-stream tasks with only minor effort for domain adaption. Correspondingly, this study tackles Camouflaged Object Detection (COD) leveraging the Segment Anything Model (SAM). The previous studies declared that SAM is not workable for COD but this study reveals that SAM works if promoted properly, for which we devise a new framework to render point promotions: First, we develop the Promotion Point Targeting Network (PPT-net) to leverage multi-scale features in predicting the probabilities of camouflaged objects' presences at given candidate points over the image. Then, we develop a key point selection (KPS) algorithm to deploy both positive and negative point promotions contrastively to SAM to guide the segmentation. It is the first work to facilitate big model for COD and achieves plausible results experimentally over the existing methods on 3 data sets under 6 metrics. This study demonstrates an off-the-shelf methodology for COD by leveraging SAM, which gains advantage over designing professional models from scratch, not only in performance, but also in turning the problem to a less challenging task, that is, seeking informative but not exactly precise promotions.
Paper Structure (18 sections, 6 equations, 3 figures, 5 tables, 1 algorithm)

This paper contains 18 sections, 6 equations, 3 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Camouflaged Object Detection (COD)
  4. Segment Anything Model (SAM)
  5. Method
  6. Visual Prompt
  7. Promotion Point Targeting Network (PPT-net)
  8. Key Point Selection (KPS) Algorithm
  9. Camouflaged Object Detection
  10. Experiments
  11. Experimental Setup
  12. Comparison with Baselines
  13. Ablation Study
  14. With/without Promotion
  15. Grid Partition
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Example of prompt.$\star$ represents Positive Point and $\star$ represents Negative Point
  • Figure 2: Overall Framework. We generate prompt points based on the input image. SAM generates a camouflaged object segmentation mask according to the image and prompt. The weights of SAM are frozen.
  • Figure 3: Overall architecture of PPT-net. It consists of two key components: The encoder to extract multi-scale features, and the predictor to output the probability of the presence of camouflaged targets at each given point.