Research map / Architectures and efficiency / CNN design
Deep Quantization: Encoding Convolutional Activations With Deep Generative Model
Zhaofan Qiu, Ting Yao, Tao Mei
Check 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 GMM Fisher Vector quantization with FV-VAE | VGG_19 pool5 representations · UCF101 split1 | accuracy: 78.85% -> 83.45% | Improved |
| Add classification loss to FV-VAE training (ablation) | VGG_19 pool5 · UCF101 split1 | accuracy: 81.91% -> 83.45% | Improved |
| Use ResNet_152 res5c activations instead of VGG_19 pool5 (ablation) | FV-VAE · UCF101 split1 | accuracy: 83.45% -> 86.33% | Improved |
| Fine-tune VGG_19 on UCF101 before FV-VAE encoding (ablation) | FV-VAE · UCF101 split1 | accuracy: 83.45% -> 82.13% | Hurt |
| Combine frame and optical-flow FV-VAE representations (ablation) | FV-VAE-pool5 · UCF101 (3 splits, ×10 augmentation) | accuracy: 83.9% -> 93.7% | Improved |
| Use ResNet_152 res5c in the fused frame and optical-flow system (ablation) | FV-VAE with pool5 optical flow · UCF101 (3 splits, ×10 augmentation) | accuracy: 93.7% -> 94.2% | Improved |
| Add IDT features to the fused FV-VAE system (ablation) | FV-VAE-(res5c + pool5 optical flow) · UCF101 (3 splits, ×10 augmentation) | accuracy: 94.2% -> 95.2% | Improved |
| Replace VGG_19 pool5 GA with FV-VAE-res5c | ActivityNet v1.3 validation · ActivityNet v1.3 validation | mAP: 76.56% -> 84.09% | Improved |
| Add VGG_19 fine-tuning on CUB-200-2011 (ablation) | FV-VAE · CUB-200-2011 test, w/o fine-tuning | accuracy: 79.3% -> 82.4% | Improved |
| Add bounding-box information at training and test time (ablation) | FV-VAE with fine-tuned VGG_19 · CUB-200-2011 test | accuracy: 82.4% -> 83.6% | Improved |
| Increase representation dimension (ablation) | FV-VAE · UCF101 split1 | accuracy | Improved |
| Vary the classification-loss weight λ₃ (ablation) | FV-VAE · UCF101 split1 | accuracy | Mixed |
Closest papers by meaning
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.87)
- MAE: Mutual Posterior-Divergence Regularization for Variational AutoEncoders (ICLR 2019 · similarity 0.87)
- Adversarial Variational Bayes: Unifying Variational Autoencoders and Generative Adversarial Networks (ICML 2017 · similarity 0.87)
- Neural Discrete Representation Learning (NeurIPS 2017 · similarity 0.87)
- SQ-VAE: Variational Bayes on Discrete Representation with Self-annealed Stochastic Quantization (ICML 2022 · similarity 0.87)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.87)
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.
