Research map / Label-efficient and robust learning / Recognition and long-tail
Inverting Visual Representations With Convolutional Networks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train an up-convolutional network to minimize image reconstruction error instead of using existing HOG inversion methods | HOG reconstruction · ImageNet validation subset | normalized reconstruction error | Helped |
| Use autoencoder training that fine-tunes the AlexNet encoder, rather than keeping AlexNet fixed (ablation) | AlexNet feature inversion · ImageNet validation subset | normalized reconstruction error | Helped |
| Binarize feature values while preserving their signs and vector norm (ablation) | AlexNet feature inversion, especially fc6 and fc7 · ImageNet validation subset | reconstruction quality | No clear effect |
| Drop out 50% of feature entries at random and renormalize (ablation) | AlexNet feature inversion · ImageNet validation subset | reconstruction accuracy | Hurt |
| Drop the 50% least non-zero activations instead of 50% random activations, then binarize (ablation) | AlexNet feature inversion, especially fc6 and fc7 · ImageNet validation subset | reconstruction error | Helped |
| Remove the five largest fc8 activations while retaining the other class probabilities (ablation) | AlexNet fc8 feature inversion · Colored apple case study | reconstruction quality | No clear effect |
| Remove all fc8 activations except the five largest (ablation) | AlexNet fc8 feature inversion · Colored apple case study | reconstruction quality | Hurt |
| Remove stride 4 from AlexNet's first layer (ablation) | AlexNet-based autoencoder · ImageNet validation subset | reconstruction error | Helped |
| Reconstruct from higher AlexNet layers rather than lower convolutional layers (ablation) | Fixed AlexNet feature inversion · ImageNet validation subset | normalized reconstruction error | Hurt |
Closest papers by meaning
- i-RevNet: Deep Invertible Networks (ICLR 2018 · similarity 0.88)
- Invertibility of Convolutional Generative Networks from Partial Measurements (NeurIPS 2018 · similarity 0.88)
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.87)
- Network Deconvolution (ICLR 2020 · similarity 0.87)
- Network Inversion of Convolutional Neural Nets (Student Abstract) (AAAI 2025 · similarity 0.86)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 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.