Out-of-distribution (OoD) semantic segmentation aims to recognize pixels of classes undefined in the training dataset. Existing methods mostly focus on training the model to fit real OoD data samples to identify OoD pixels, which requires extra data collection and annotation efforts. By contrast, synthesizing OoD data with training data provides a more resource-efficient alternative. However, synthetic data generated from controlled settings lacks diversity, causing the model to suffer from overfitting. To this end, we propose a disentangled representation learning (DRL) method to guide the model to disentangle semantic-related and semantic-unrelated features from synthetic OoD data. DRL encourages the model to utilize the former to identify semantic categories, rather than overfitting to such semantic-unrelated features as synthetic artificiality. Specifically, DRL first incorporates two disentanglers to extract the semantic-related and -unrelated features and then applies a shuffle and reconstruction mechanism to regularize the disentangled features. Furthermore, to facilitate disentangling, we propose a pixel-wise feature similarity calibration (PSC) module, which utilizes more accurate ID-OoD similarity to calibrate inaccurate ID-OoD similarity learned exclusively from ID data. Thus, PSC delivers accurate and stable pixel-wise features for effective disentangling. Extensive experiments illustrate that the proposed method exhibits strong generalization ability. It attains 74.04% AuPRC and 20.82% FPR on Road Anomaly, 69.85% AuPRC and 5.78% FPR on Fishyscapes LostAndFound Validation Set, using SegFormer with the MiT-B5 backbone. Source code is available at https://github.com/WanMotion/DisentangledOoDSeg.

The transferability of adversarial examples is vital for black-box attacks, as it enables the adversary to deceive the target model without knowing its internals. Despite numerous methods focusing on transferability, they still struggle with transferring across models with distinct architectural components (e.g., CNNs and ViTs). In this work, we argue that the limited adversarial perturbation diversity leads to overfitting of the surrogate model, which acts as a key factor in reducing transferability. To this end, we propose a Masked Adversarial Perturbation (MAP) method to boost adversarial transferability across various architectures from a novel perspective of diversifying perturbation. Specifically, MAP randomly masks perturbation patches during iterations and compels the remaining ones to retain the attack effect, which diversifies perturbations to mitigate their overfitting to the surrogate model. Naturally, MAP spreads perturbation over local patches to alleviate their co-adaptation and prevent perturbations from overly relying on specific patterns. Consequently, it can deceive convolution operation and self-attention mechanism indiscriminately by attacking their basic input units, i.e., a single patch, showing superior transferability over previous methods. Extensive experiments illustrate that MAP consistently and significantly boosts diverse black-box attacks to achieve state-of-the-art performance.

Context modeling or multi-level feature fusion methods have been proved to be effective in improving semantic segmentation performance. However, they are not specialized to deal with the problems of pixel-context mismatch and spatial feature misalignment, and the high computational complexity hinders their widespread application in real-time scenarios. In this work, we propose a lightweight Context and Spatial Feature Calibration Network (CSFCN) to address the above issues with pooling-based and sampling-based attention mechanisms. CSFCN contains two core modules: Context Feature Calibration (CFC) module and Spatial Feature Calibration (SFC) module. CFC adopts a cascaded pyramid pooling module to efficiently capture nested contexts, and then aggregates private contexts for each pixel based on pixel-context similarity to realize context feature calibration. SFC splits features into multiple groups of sub-features along the channel dimension and propagates sub-features therein by the learnable sampling to achieve spatial feature calibration. Extensive experiments on the Cityscapes and CamVid datasets illustrate that our method achieves a state-of-the-art trade-off between speed and accuracy. Concretely, our method achieves 78.7% mIoU with 70.0 FPS and 77.8% mIoU with 179.2 FPS on the Cityscapes and CamVid test sets, respectively. The code is available at https://nave.vr3i.com/ and https://github.com/kaigelee/CSFCN.