Research map / Image generation / Autoregressive and latent generation
DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| fine-tune with LoRA instead of MLP head (ablation) | CLIP+OpenCLIP+DINO ensemble · NIGHTS 2AFC test | human alignment (%): 93.4 -> 96.2 | Helped |
| add MLP-head tuning on NIGHTS to pretrained ensemble (ablation) | CLIP+OpenCLIP+DINO ensemble · NIGHTS 2AFC test | human alignment (%): 90.8 -> 93.4 | Helped |
| add LoRA tuning on NIGHTS to pretrained ensemble (ablation) | CLIP+OpenCLIP+DINO ensemble · NIGHTS 2AFC test | human alignment (%): 90.8 -> 96.2 | Helped |
| concatenate features of DINO, CLIP and OpenCLIP into ensemble (with LoRA) (ablation) | LoRA-tuned single backbones · NIGHTS 2AFC test | human alignment (%): 95.5 -> 96.2 | Helped |
| LoRA tune DINO ViT B/16 (ablation) | DINO · NIGHTS 2AFC test | human alignment (%): 90.1 -> 94.6 | Helped |
| LoRA tune MAE ViT B/16 (ablation) | MAE · NIGHTS 2AFC test | human alignment (%): 82.1 -> 90.5 | Helped |
| tune on NIGHTS (low-level BAPPS evaluation) | DINO/CLIP/OpenCLIP/ensemble · BAPPS | human alignment | Helped |
| tune on NIGHTS (high-level THINGS evaluation) | large vision models · THINGS | human alignment | Hurt |
| tune on NIGHTS (IQA TID2013) | DreamSim · TID2013 | SROCC: 0.757 -> 0.814 | Helped |
| tune on NIGHTS (IQA KADID-10k) | DreamSim · KADID-10k | SROCC: 0.812 -> 0.868 | Helped |
| remove foreground vs background content from image (ablation) | DreamSim · NIGHTS image ablations | agreement with unmodified decisions | Mixed |
Closest papers by meaning
- The Many Senses of Visual Similarity: A Text-Prompted Image Perceptual Metric (NeurIPS 2026 · similarity 0.93)
- When does perceptual alignment benefit vision representations? (NeurIPS 2024 · similarity 0.88)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.88)
- LipSim: A Provably Robust Perceptual Similarity Metric (ICLR 2024 · similarity 0.88)
- PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception (ICML 2026 · similarity 0.87)
- Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding (CVPR 2024 · similarity 0.87)
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.