Research map / Detection and segmentation / CNN features
Visualizing Object Detection Features
Carl Vondrick, Aditya Khosla, Hamed Pirsiavash, Tomasz Malisiewicz, Antonio B. Torralba
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 |
|---|---|---|---|
| use paired dictionary learning instead of the standard HOG glyph | HOG visualization · MTurk classification of PASCAL VOC windows | accuracy: 0.191 -> 0.383 | Improved |
| use direct optimization instead of the HOG glyph | HOG visualization · MTurk classification of PASCAL VOC windows | accuracy: 0.191 -> 0.355 | Improved |
| use exemplar LDA instead of paired dictionary | HOG inversion · Pixel-level reconstruction (NCC) | mean NCC: 0.637 -> 0.671 | Improved |
| use exemplar LDA instead of paired dictionary | HOG inversion · MTurk classification of inversions | accuracy: 0.383 -> 0.282 | Hurt |
| use ridge regression instead of exemplar LDA | HOG inversion · Pixel-level reconstruction (NCC) | mean NCC: 0.671 -> 0.656 | Hurt |
| use expert PhD students instead of MTurk workers on glyphs | HOG glyph · MTurk-style classification | accuracy: 0.191 -> 0.233 | Improved |
| experts viewing paired dictionary instead of non-experts | HOG visualization · classification | accuracy | No clear effect |
| use glyph for bicycles instead of paired dictionary | HOG visualization · MTurk classification, bicycle | accuracy: 0.307 -> 0.405 | Improved |
| use RGB patches instead of HOG visualization for human chair detection | Human sliding-window detector · PASCAL VOC chair detection | accuracy | Improved |
| add color-averaging affinity diversity cost to multiple inversions (ablation) | paired dictionary on CNN pool5 features · Multiple inversions, VOC 2007 windows | image distance at fixed feature distance | Improved |
| add edge affinity diversity cost to multiple inversions (ablation) | paired dictionary on CNN pool5 features · Multiple inversions, VOC 2007 windows | image distance at fixed feature distance | Improved |
| use nudged or subset dictionaries instead of diversity cost (ablation) | multiple inversions baseline · Multiple inversions, VOC 2007 windows | diversity vs feature match | Hurt |
Closest papers by meaning
- Black-Box Explanation of Object Detectors via Saliency Maps (CVPR 2021 · similarity 0.86)
- Exploring the Bounds of the Utility of Context for Object Detection (CVPR 2019 · similarity 0.85)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.85)
- An Introduction to Rule-Based Feature and Object Perception for Middle School Students (AAAI 2023 · similarity 0.85)
- Explaining Object Detectors via Collective Contribution of Pixels (CVPR 2026 · similarity 0.84)
- The Devil is in the Fine-Grained Details: Evaluating Open-Vocabulary Object Detectors for Fine-Grained Understanding (CVPR 2024 · similarity 0.83)
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.
