Introduces a semantic hierarchy-aware progressive codec that decomposes latents into ordered channel blocks for improved coarse recognition at low bitrates while preserving fine-grained accuracy at higher rates.
hub
Deep residual learning for image recognition
10 Pith papers cite this work. Polarity classification is still indexing.
10
Pith papers citing it
hub tools
citation-role summary
method 1
citation-polarity summary
roles
method 1polarities
use method 1representative citing papers
HAAD detects deepfakes by modeling latent manifolds as potential energy surfaces and quantifying instability via Hamiltonian trajectory statistics such as action and energy dissipation.
CAAP creates universal cross-shaped adversarial patches that disrupt palmprint recognition models under realistic capture distortions, showing high attack success and partial resistance to adversarial training on multiple datasets.
Coward detects backdoors in federated learning by injecting a collision-suppressed watermark on OOD data to invert the detection paradigm and limit OOD bias effects.
DFBScanner detects backdoors by combining anomaly indicators from final-layer parameters into a Trojan clue score, reporting 97.17% true-positive rate, 0.95% false-positive rate, and 1 ms average detection time on a benchmark of over 5,000 models.
CORF unifies domain generalization and class-incremental learning via selective sample refinement with spatial maps and confidence weighting plus cascaded relational distillation.
BicKD introduces a bilateral contrastive loss in knowledge distillation that strengthens class-wise orthogonality and intra-class consistency in predictive distributions, outperforming prior logit-based methods.
DualGeo improves worldwide image geo-localization by fusing visual and semantic features with dual-view contrastive learning and refining retrievals via geographic clustering plus LMMs, achieving 3.6-16.58% better street-level and 1.29-8.77% better city-level accuracy on standard benchmarks.
Catastrophic overfitting in fast adversarial training is reinterpreted as a weak-trigger variant of unlearnable tasks, allowing backdoor-inspired recalibration and outlier suppression to restore robustness.
The paper derives closed-form minimum achievable rates under semantic distance and complexity constraints for Gaussian and binary sources, demonstrating a fundamental three-way tradeoff validated on image and video data.
citing papers explorer
-
Coarse-to-Fine: Progressive Image Compression for Semantically Hierarchical Classification
Introduces a semantic hierarchy-aware progressive codec that decomposes latents into ordered channel blocks for improved coarse recognition at low bitrates while preserving fine-grained accuracy at higher rates.
-
Detecting Deepfakes via Hamiltonian Dynamics
HAAD detects deepfakes by modeling latent manifolds as potential energy surfaces and quantifying instability via Hamiltonian trajectory statistics such as action and energy dissipation.
-
CAAP: Capture-Aware Adversarial Patch Attacks on Palmprint Recognition Models
CAAP creates universal cross-shaped adversarial patches that disrupt palmprint recognition models under realistic capture distortions, showing high attack success and partial resistance to adversarial training on multiple datasets.
-
Coward: Collision-based OOD Watermarking for Practical Proactive Federated Backdoor Detection
Coward detects backdoors in federated learning by injecting a collision-suppressed watermark on OOD data to invert the detection paradigm and limit OOD bias effects.
-
Lightweight and Fast Backdoor Model Detection
DFBScanner detects backdoors by combining anomaly indicators from final-layer parameters into a Trojan clue score, reporting 97.17% true-positive rate, 0.95% false-positive rate, and 1 ms average detection time on a benchmark of over 5,000 models.
-
Cross-Sample Relational Fusion: Unifying Domain Generalization and Class-Incremental Learning
CORF unifies domain generalization and class-incremental learning via selective sample refinement with spatial maps and confidence weighting plus cascaded relational distillation.
-
BicKD: Bilateral Contrastive Knowledge Distillation
BicKD introduces a bilateral contrastive loss in knowledge distillation that strengthens class-wise orthogonality and intra-class consistency in predictive distributions, outperforming prior logit-based methods.
-
DualGeo: A Dual-View Framework for Worldwide Image Geo-localization
DualGeo improves worldwide image geo-localization by fusing visual and semantic features with dual-view contrastive learning and refining retrievals via geographic clustering plus LMMs, achieving 3.6-16.58% better street-level and 1.29-8.77% better city-level accuracy on standard benchmarks.
-
Unveiling the Backdoor Mechanism Hidden Behind Catastrophic Overfitting in Fast Adversarial Training
Catastrophic overfitting in fast adversarial training is reinterpreted as a weak-trigger variant of unlearnable tasks, allowing backdoor-inspired recalibration and outlier suppression to restore robustness.
-
On the Rate-Distortion-Complexity Tradeoff for Semantic Communication
The paper derives closed-form minimum achievable rates under semantic distance and complexity constraints for Gaussian and binary sources, demonstrating a fundamental three-way tradeoff validated on image and video data.