Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
"Contributions of Shape, Texture, and Color in Visual Recognition"
Yunhao Ge, Yao Xiao, Zhi Xu, Xingrui Wang, Laurent Itti
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 |
|---|---|---|---|
| combine shape, texture and color nets instead of shape net alone (ablation) | Humanoid NN · Shape-biased subset | accuracy: 90% -> 95% | Improved |
| combine three feature nets instead of texture net alone (ablation) | Humanoid NN · Texture-biased subset | accuracy: 81% -> 91% | Improved |
| combine three feature nets instead of color net alone (ablation) | Humanoid NN · Color-biased subset | accuracy: 82% -> 92% | Improved |
| decompose image into three feature nets instead of original ResNet-18 (ablation) | Humanoid NN · Color-biased subset | accuracy: 95% -> 92% | Hurt |
| add color to shape+texture inputs (ablation) | Humanoid NN · Color-biased subset | accuracy: 83% -> 92% | Improved |
| add shape to texture+color inputs (ablation) | Humanoid NN · Shape-biased subset | accuracy: 86% -> 95% | Improved |
| add shape to texture+color inputs (ablation) | Humanoid NN · Texture-biased subset | accuracy: 83% -> 90% | Improved |
| replace MLP aggregation with average pooling (ablation) | Humanoid NN · Color-biased subset | color contribution ratio: 70% -> 95% | Mixed |
| replace MLP aggregation with average pooling (ablation) | Humanoid NN · Texture-biased subset | texture contribution ratio: 65% -> 79% | Mixed |
| vary random seed of texture extractor (ablation) | Humanoid NN · Texture-biased subset | texture contribution ratio: 62.4% ± 2.3% | No clear effect |
| use HVE cross-feature GAN instead of three pix2pix GANs | imagination · color input | FID: 203.527 -> 52.846 | Improved |
| use HVE cross-feature GAN instead of three pix2pix GANs | imagination · shape input | FID: 123.915 -> 96.871 | Improved |
Closest papers by meaning
- Visual Anagrams Reveal Hidden Differences in Holistic Shape Processing Across Vision Models (NeurIPS 2025 · similarity 0.89)
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.88)
- Hidden in Plain Sight: Evaluating Abstract Shape Recognition in Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Harmonizing the object recognition strategies of deep neural networks with humans (NeurIPS 2022 · similarity 0.86)
- Intriguing Properties of Generative Classifiers (ICLR 2024 · similarity 0.85)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 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.
