Research map / Vision-language and multimodal / Multimodal LLMs
ECC: Encoder-Centric Corruption for Fine-Grained Vision in VLMs
Hyesong Choi, Daeun Kim, Sungmin Cha, Kwang Moo Yi, Dongbo Min
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 Conjunctive instead of Substitutive Corruption (ablation) | noise-based C2R pretraining · FGVC and recognition benchmarks | transfer accuracy | Improved |
| move corruption and reconstruction into the encoder (Encoder-style) vs Decoder-style (ablation) | naive noise-based C2R · recognition and FGVC | transfer accuracy | Improved |
| inject noise at feature level (blocks 2,4,6) instead of pixel level (block 0) (ablation) | ECC encoder · recognition tasks | transfer accuracy | Improved |
| add disruption loss to separate mask reconstruction and denoising (ablation) | ECC · fine-grained and standard recognition | accuracy | Improved |
| add noise-based denoising to MIM (DiffMAE, MaskDiT vs MAE) | MAE · FGVC (CUB) | top-1 accuracy: 79.14 -> 78.50 | Hurt |
| use ECC instead of MAE | MAE · Aircraft | top-1 accuracy: 81.56 -> 84.42 | Improved |
| use ECC instead of MAE | MAE · COCO (mask) | AP: 41.74 -> 47.90 | Improved |
| apply ECC to CLIP vision encoder | CLIP ViT-B/16 · fine-tuning | accuracy: 83.35 -> 84.83 | Improved |
| apply ECC to CLIP vision encoder | CLIP · visual grounding | attention entropy (bits): 4.2 -> 3.5 | Improved |
| apply ECC to LLaVA-1.5 vision encoder | LLaVA-1.5 · object hallucination | CHAIRs: 48.4 -> 46.5 | Improved |
| apply ECC to LLaVA-1.5 vision encoder | LLaVA-1.5 · Flickr30K | R@1: 69.1 -> 70.2 | Improved |
Closest papers by meaning
- Mint: A Simple Test-Time Adaptation of Vision-Language Models against Common Corruptions (NeurIPS 2025 · similarity 0.88)
- Revisiting Visual Corruptions in LVLMs: A Shape-Texture Perspective on Model Failures (CVPR 2026 · similarity 0.87)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.85)
- Beyond Encoder Accumulation: Measuring Encoder Roles in Multi-Encoder VLMs (NeurIPS 2026 · similarity 0.85)
- Benchmarking and Enhancing VLM for Compressed Image Understanding (ICML 2026 · similarity 0.85)
- LORE: Lagrangian-Optimized Robust Embeddings for Visual Encoders (NeurIPS 2025 · similarity 0.84)
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.
