Research map / Vision-language and multimodal / Human and embodied modeling
Visual Concepts Tokenization
Tao Yang, Yuwang Wang, Yan Lu, Nanning Zheng
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 pretrained AE instead of random AE as tokenizer (ablation) | VCT · Shapes3D ablation | MIG: 0.484 -> 0.560 | Improved |
| use patch tokenizer instead of AE (ablation) | VCT · Shapes3D ablation | MIG: 0.484 -> 0.361 | Hurt |
| use pretrained VQ-VAE instead of AE (ablation) | VCT · Shapes3D ablation | DCI: 0.802 -> 0.884 | Improved |
| remove concept disentangling loss (ablation) | AE + VCT · Shapes3D ablation | MIG: 0.484 -> 0.165 | Hurt |
| remove concept disentangling loss (ablation) | pretrained VQVAE + VCT · Shapes3D ablation | MIG: 0.525 -> 0.286 | Hurt |
| add self-attention after cross-attention in Concept Tokenizer (ablation) | VCT · Shapes3D ablation | MIG: 0.525 -> 0.000 | Hurt |
| remove detach (ablation) | VCT · Shapes3D ablation | MIG: 0.525 -> 0.392 | Hurt |
| add positional embedding (ablation) | VCT · Shapes3D ablation | MIG: 0.525 -> 0.525 | No clear effect |
| replace Concept Detokenizer with CNN (ablation) | VCT · Shapes3D ablation | MIG: 0.525 -> 0.157 | Hurt |
| replace Concept Detokenizer with Transformer (ablation) | VCT · Shapes3D ablation | MIG: 0.525 -> 0.467 | Hurt |
| increase batch size from 32 to 64 (ablation) | VCT · Shapes3D ablation | DCI: 0.884 -> 0.900 | Improved |
| decrease batch size from 32 to 16 (ablation) | VCT · Shapes3D ablation | DCI: 0.884 -> 0.862 | Hurt |
Closest papers by meaning
- A More Word-like Image Tokenization for MLLMs (CVPR 2026 · similarity 0.88)
- Discovering Fine-Grained Visual-Concept Relations by Disentangled Optimal Transport Concept Bottleneck Models (CVPR 2025 · similarity 0.88)
- ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features (ICML 2025 · similarity 0.87)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.87)
- Attention Calibration for Disentangled Text-to-Image Personalization (CVPR 2024 · similarity 0.87)
- Revisiting Disentanglement in Downstream Tasks: A Study on Its Necessity for Abstract Visual Reasoning (AAAI 2024 · similarity 0.86)
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.
