Research map / Image generation / Autoregressive and latent generation
LOTA: Bit-Planes Guided AI-Generated Image Detection
Hongsong Wang, Renxi Cheng, Yang Zhang, Chaolei Han, Jie Gui
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 bit-plane low-bit noise images instead of raw images (BGNIG) (ablation) | LOTA-ngc · GenImage (train SD V1.5) | average accuracy (%): 87.5 -> 96.8 | Improved |
| add maximum gradient patch selection (MGPS) (ablation) | LOTA-ngc · GenImage (train SD V1.5) | average accuracy (%): 90.9 -> 96.8 | Improved |
| use NBC/NGC classifier instead of a fully connected layer (ablation) | LOTA · GenImage (train SD V1.5) | average accuracy (%): 52.6 -> 96.8 | Improved |
| select highest-score patch instead of lowest-score (ablation) | MGPS · GenImage (train SD V1.5) | average accuracy (%): 89.6 -> 96.8 | Improved |
| select highest-score patch instead of random patch (ablation) | MGPS · GenImage (train SD V1.5) | average accuracy (%): 93.1 -> 96.8 | Improved |
| use 32x32 patch instead of 16x16, 48x48 or 64x64 (ablation) | MGPS · GenImage (train SD V1.5) | average accuracy (%): 92.7 / 91.4 / 90.8 -> 96.8 | Improved |
| compose the three lowest bit-planes instead of fewer or more (ablation) | BGNIG · GenImage (train SD V1.5) | average accuracy (%): 91.5 (bit 0) -> 96.8 (bits 0~2); 86.4 at bits 0~5 | Improved |
| use thresholding instead of scaling normalization (ablation) | LOTA · GenImage (avg) | average accuracy (%): 98.7 -> 98.9 | No clear effect |
| use noise-based instead of noise-guided classifier (ablation) | LOTA · GenImage (avg) | average accuracy (%): 93.2 -> 98.9 | Improved |
| replace multi-step reconstruction error with bit-plane noise for error extraction | LOTA vs ESSP · GenImage (avg) | average accuracy (%): 87.0 -> 98.9 | Improved |
| replace reconstruction-based error extraction with bit-plane extraction | LOTA · error extraction time | time per image: 25.10 ms -> 1.52 ms (vs ESSP) | Improved |
| apply Gaussian blur and JPEG compression at test time | LOTA vs LaRE2 and ESSP · GenImage (train SD V1.5) | accuracy under degradation | Improved |
Later papers that compared with this method
1 later accepted paper reports LOTA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- PPM-CLIP: Probabilistic Prompt Modeling for Generalizable AI-Generated Image Detection (CVPR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Beyond Generation: A Diffusion-based Low-level Feature Extractor for Detecting AI-generated Images (CVPR 2025 · similarity 0.87)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.85)
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.85)
- Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · similarity 0.84)
- Training-free Detection of AI-generated images via Cropping Robustness (NeurIPS 2025 · similarity 0.84)
- FerretNet: Efficient Synthetic Image Detection via Local Pixel Dependencies (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.
