Research map / Detection and segmentation / CNN features
Guided Perturbations: Self-Corrective Behavior in Convolutional Neural Networks
Swami Sankaranarayanan, Arpit Jain, Ser Nam Lim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add guided perturbations at test time | FCN-32s · VOC2012 reduced val (346 images) | mIoU: 62.10 -> 64.71 | Improved |
| add guided perturbations at test time | FCN-8s · VOC2012 reduced val (346 images) | mIoU: 63.97 -> 66.97 | Improved |
| add guided perturbations at test time | FCN-8s-coco · VOC2012 reduced val (346 images) | mIoU: 69.85 -> 71.99 | Improved |
| add guided perturbations at test time | CRFasRNN-coco · VOC2012 reduced val (346 images) | mIoU: 72.95 -> 73.75 | Improved |
| add guided perturbations at test time | Deeplab-VGG16 · VOC2012 reduced val (346 images) | mIoU: 66.9 -> 69.1 | Improved |
| add guided perturbations at test time | Deeplab-ResNet101 · VOC2012 reduced val (346 images) | mIoU: 74.1 -> 75.3 | Improved |
| add guided perturbations at test time | FCN-8s · PASCAL-Context 59 classes val | mean IU: 39.12 -> 40.44 | Improved |
| add guided perturbations at test time | FCN-8s · VOC2012 test | mean IoU: 67.2 -> 69.3 | Improved |
| add guided perturbations using nearest-neighbor class gradients | small CNN · MNIST | accuracy: 98.92 -> 99.15 | Improved |
| add guided perturbations using nearest-neighbor class gradients | small CNN · CIFAR10 | accuracy: 76.31 -> 76.95 | Improved |
| add GP on top of CRF post-processing (ablation) | FCN-8s-coco + CRF · VOC2012 reduced val | mIoU: 71.1 -> 72.7 | Improved |
| truncate perturbation backpropagation to pool4 instead of input (ablation) | FCN-32s + GP · VOC2012 reduced val | mIoU and time: 64.71 -> 64.3 mIoU, 0.33s -> 0.22s | Mixed |
Closest papers by meaning
- Perturbative Neural Networks (CVPR 2018 · similarity 0.91)
- Deep Perturbation Learning: Enhancing the Network Performance via Image Perturbations (ICML 2023 · similarity 0.88)
- Universal Adversarial Perturbations Against Semantic Image Segmentation (ICCV 2017 · similarity 0.87)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.86)
- Towards the Uncharted: Density-Descending Feature Perturbation for Semi-supervised Semantic Segmentation (CVPR 2024 · similarity 0.85)
- MRFP: Learning Generalizable Semantic Segmentation from Sim-2-Real with Multi-Resolution Feature Perturbation (CVPR 2024 · similarity 0.85)
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.
