Research map / Label-efficient and robust learning / Recognition and long-tail
Interpretable Image Recognition by Constructing Transparent Embedding Space
Jiaqi Wang, Huafeng Liu, Xinyue Wang, Liping Jing
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 |
|---|---|---|---|
| add orthonormality loss Lorth to Lid+Lcs (ablation) | TesNet VGG16/VGG19/ResNet34/DenseNet121/DenseNet161 · CUB-200-2011 (cropped, augmented) | accuracy (%): VGG16 79.2 -> 79.4; VGG19 79.8 -> 80.3; ResNet34 80.7 -> 81.1; Dense121 83.4 -> 84.1; Dense161 82.6 -> 83.0 | Improved |
| add orthonormality loss Lorth to Lid+Lcs (ablation) | TesNet ResNet152 · CUB-200-2011 (cropped, augmented) | accuracy (%): 81.6 -> 80.3 | Hurt |
| add subspace separation loss Lss to Lid+Lcs+Lorth (ablation) | TesNet VGG16 · CUB-200-2011 (cropped, augmented) | accuracy (%): 79.4 -> 81.3 | Improved |
| add subspace separation loss Lss to Lid+Lcs+Lorth (ablation) | TesNet ResNet152 · CUB-200-2011 (cropped, augmented) | accuracy (%): 80.3 -> 82.7 | Improved |
| add subspace separation loss Lss to Lid+Lcs+Lorth (ablation) | TesNet DenseNet121 · CUB-200-2011 (cropped, augmented) | accuracy (%): 84.1 -> 84.8 | Improved |
| add subspace separation loss Lss to Lid+Lcs+Lorth (ablation) | TesNet ResNet34 · CUB-200-2011 (cropped, augmented) | accuracy (%): 81.1 -> 82.8 | Improved |
| replace ProtoPNet with TesNet | ProtoPNet VGG16 · CUB-200-2011 (cropped, augmented) | accuracy (%): 77.2 -> 81.3 | Improved |
| replace ProtoPNet with TesNet | ProtoPNet DenseNet121 · CUB-200-2011 (cropped, augmented) | accuracy (%): 79.0 -> 84.8 | Improved |
| switch from non-interpretable baseline to TesNet | VGG16 · CUB-200-2011 (cropped, augmented) | accuracy (%): 73.3 -> 81.3 | Improved |
| switch from non-interpretable baseline to TesNet | ResNet152 · Stanford Cars (cropped, augmented) | accuracy (%): 92.8 -> 92.0 | Hurt |
| switch from non-interpretable baseline to TesNet | ResNet34 · Stanford Cars (cropped, augmented) | accuracy (%): 92.6 -> 90.9 | Hurt |
| replace ProtoPNet with TesNet | ProtoPNet ResNet34 · Stanford Cars (cropped, augmented) | accuracy (%): 88.8 -> 90.9 | Improved |
Closest papers by meaning
- Enhancing Pre-trained Representation Classifiability can Boost its Interpretability (ICLR 2025 · similarity 0.84)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.84)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.84)
- NeSyFOLD: A Framework for Interpretable Image Classification (AAAI 2024 · similarity 0.83)
- Enhancing Performance of Explainable AI Models with Constrained Concept Refinement (ICML 2025 · similarity 0.83)
- A Simple Interpretable Transformer for Fine-Grained Image Classification and Analysis (ICLR 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.
