Research map / Label-efficient and robust learning / OOD and anomaly detection
Skip-GANomaly++: Skip Connections and Residual Blocks for Anomaly Detection (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Skip-GANomaly++, a semi-supervised GAN-based anomaly detector combining skip connections, residual units, and ASPP, evaluated one-class-out on CIFAR-10 against GANomaly, Skip-GANomaly, AnoGAN, and EGBAD.
Problem
Anomaly detection in images without labeled abnormal data, where supervised approaches need large labeled sets; prior GAN-based detectors like GANomaly and Skip-GANomaly have generalization issues from over-emphasized features.
Approach
Extend Skip-GANomaly with residual units in every block and ASPP at the encoder-decoder junction, so skip connections are stabilized and multi-scale context is preserved, while training only on normal data and scoring by reconstruction similarity.
Key results
- Proposed model achieves AUC 0.998 on truck and cat and 0.999 on deer, versus at most 0.931 for the best baseline on those classes.
- Proposed model underperforms the baselines on 'car' and 'truck' per the authors' statement, while outperforming on the other classes; e.g., bird 0.891 vs 0.631 for GANomaly (best baseline on that class among those listed in Table 1)
- Proposed model outperforms baselines on most CIFAR-10 one-class settings, with notable gains on bird, deer, and cat.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 bird anomaly detection · CIFAR-10 bird | AUC: 0.953 -> 0.891 | Hurt |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 car anomaly detection · CIFAR-10 car | AUC: 0.907 -> 0.892 | Hurt |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 truck anomaly detection · CIFAR-10 truck | AUC: 0.659 -> 0.998 | Helped |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 ship anomaly detection · CIFAR-10 ship | AUC: 0.607 -> 0.907 | Helped |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 cat anomaly detection · CIFAR-10 cat | AUC: 0.602 -> 0.998 | Helped |
| Use Skip-GANomaly++ instead of GANomaly | CIFAR-10 deer anomaly detection · CIFAR-10 deer | AUC: 0.683 -> 0.999 | Helped |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 frog anomaly detection · CIFAR-10 frog | AUC: 0.788 -> 0.832 | Helped |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 horse anomaly detection · CIFAR-10 horse | AUC: 0.797 -> 0.984 | Helped |
| Use Skip-GANomaly++ instead of Skip-GANomaly | CIFAR-10 airplane anomaly detection · CIFAR-10 airplane | AUC: 0.615 -> 0.818 | Helped |
| Use Skip-GANomaly++ instead of GANomaly | CIFAR-10 dog anomaly detection · CIFAR-10 dog | AUC: 0.628 -> 0.818 | Helped |
Surprising findings
- Proposed model is reported to underperform on 'car' and 'truck' classes, though Table 1 shows truck at 0.998 for the proposed model versus 0.659 for Skip-GANomaly, which conflicts with the text.
Limitations
- Evaluated only on CIFAR-10 at 32x32 resolution as a proof of concept; no medical data despite motivation.
- Comparison limited to four older baselines; no statistical variance or seeds reported.
- Claims of 'more efficient' are not supported by reported measurements.
Experimental setup
Datasets: CIFAR-10
Benchmarks: CIFAR-10
Metrics: AUC-ROC
Baselines: GANomaly Skip-GANomaly AnoGAN EGBAD
Builds on: GANomaly Skip-GANomaly ResNet ASPP
Closest papers by meaning
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.84)
- Normal-Abnormal Guided Generalist Anomaly Detection (NeurIPS 2025 · similarity 0.84)
- Toward Generalist Anomaly Detection via In-context Residual Learning with Few-shot Sample Prompts (CVPR 2024 · similarity 0.83)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.83)
- Vague Prototype-Oriented Diffusion Model for Multi-Class Anomaly Detection (ICML 2024 · similarity 0.83)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.