Research map / Adversarial robustness and security / Certified and empirical robustness
Learning Discriminative Video Representations Using Adversarial Perturbations
Jue Wang, Anoop Cherian
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 |
|---|---|---|---|
| replace average pooling with discriminative subspace pooling | two-stream ResNet-152 · HMDB-51 split-1 | accuracy: 63.8% -> 72.5% | Improved |
| replace max pooling with discriminative subspace pooling | two-stream ResNet-152 · HMDB-51 split-1 | accuracy: 60.6% -> 72.5% | Improved |
| replace average pooling with discriminative subspace pooling | Res-TCN skeleton features · NTU-RGBD cross-subject | accuracy: 74.3% -> 81.6% | Improved |
| replace average pooling with discriminative subspace pooling | Inception-ResNet-v2 · YUP++ moving | accuracy: 76.5% -> 88.3% | Improved |
| use universal adversarial perturbation noise instead of Gaussian noise as negative bag (ablation) | DSP · HMDB-51 split-1 | accuracy: 69.8% -> 70.8% | Improved |
| increase number of hyperplanes from one to six (ablation) | DSP · HMDB-51 split-1 | accuracy: about 3% accuracy gain | Improved |
| increase number of hyperplanes from six to fifteen (ablation) | DSP · HMDB-51 split-1 | accuracy: drops around 0.5% | No clear effect |
| add temporal ordering constraints (ablation) | DSP · HMDB-51 split-1 | accuracy | Improved |
| increase UAP fooling rate from 0.1 to 0.9 (ablation) | DSP · HMDB-51 split-1 | accuracy | Improved |
| apply DSP on I3D features instead of average pooling | TS I3D · HMDB-51 | accuracy: 80.9% -> 81.5% | Improved |
| apply DSP instead of average pooling on I3D, long sequences | I3D · HMDB-51 split-1 (260+ frames) | accuracy: 77.3 -> 82.6 | Improved |
| apply DSP instead of average pooling on I3D, short sequences | I3D · HMDB-51 split-1 (1+ frames) | accuracy: 80.8 -> 81.6 | Improved |
Closest papers by meaning
- Discriminative Video Representation Learning Using Support Vector Classifiers (TPAMI 2019 · similarity 0.88)
- Video Representation Learning Using Discriminative Pooling (CVPR 2018 · similarity 0.87)
- Kernelized Subspace Pooling for Deep Local Descriptors (CVPR 2018 · similarity 0.87)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.86)
- LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders (NeurIPS 2025 · similarity 0.85)
- Exploring the Precise Dynamics of Single-Layer GAN Models: Leveraging Multi-Feature Discriminators for High-Dimensional Subspace Learning (NeurIPS 2024 · similarity 0.84)
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.
