Uncertainty Aware Human-machine Collaboration in Camouflaged Object Detection

TL;DR

This work introduces an uncertainty-aware human–machine collaboration framework for camouflaged object detection (COD) that fuses a multiview CV backbone with RSVP-based BCIs. During training, uncertainty-driven policies selectively augment and sample high- versus low-confidence images; during testing, low-confidence cases are redirected to an EEG-based RSVP system to improve decision reliability. On CAMO, the approach achieves state-of-the-art performance, with average improvements of BA and F1 by 4.56% and 3.66%, and up to 7.6% BA and 6.66% F1 for top participants. The framework reduces cognitive effort by deferring only uncertain cases to human operators, demonstrating a robust, scalable path toward trustworthy COD in real-world scenarios.

Abstract

Camouflaged Object Detection (COD), the task of identifying objects concealed within their environments, has seen rapid growth due to its wide range of practical applications. A key step toward developing trustworthy COD systems is the estimation and effective utilization of uncertainty. In this work, we propose a human-machine collaboration framework for classifying the presence of camouflaged objects, leveraging the complementary strengths of computer vision (CV) models and noninvasive brain-computer interfaces (BCIs). Our approach introduces a multiview backbone to estimate uncertainty in CV model predictions, utilizes this uncertainty during training to improve efficiency, and defers low-confidence cases to human evaluation via RSVP-based BCIs during testing for more reliable decision-making. We evaluated the framework in the CAMO dataset, achieving state-of-the-art results with an average improvement of 4.56\% in balanced accuracy (BA) and 3.66\% in the F1 score compared to existing methods. For the best-performing participants, the improvements reached 7.6\% in BA and 6.66\% in the F1 score. Analysis of the training process revealed a strong correlation between our confidence measures and precision, while an ablation study confirmed the effectiveness of the proposed training policy and the human-machine collaboration strategy. In general, this work reduces human cognitive load, improves system reliability, and provides a strong foundation for advancements in real-world COD applications and human-computer interaction. Our code and data are available at: https://github.com/ziyuey/Uncertainty-aware-human-machine-collaboration-in-camouflaged-object-identification.

Figures (3)

• Figure 1: Overall framework of this study. During training, the dataset is split into high and low-confidence sets, with augmentations applied to low-confidence samples and the split updated per epoch; during testing, high-uncertainty samples are classified using the RSVP program, while high-confidence samples are handled by the CV model, enhancing COD performance.
• Figure 2: RSVP Paradigm Design. At the start of each trial, a fixation cross appeared in the center of the screen. The stimuli were then presented at a frequency of 1 Hz. The participants were given ample rest time between blocks. The experiment ensured that there were at least three nontarget images between any two target images. All stimulus images were displayed on a 512 × 512 resolution monitor with a refresh rate of 60 Hz.
• Figure 3: Examples of CV model failure cases where the CV model incorrectly identified background as containing camouflaged objects.