Update README.md

README.md

@@ -13,7 +13,7 @@ task_categories:
   - other
 size_categories:
   - 1K<n<10K
-homepage: https://huggingface.co/datasets/eoguzakin/aigi-inpainting-robustness
+homepage: https://huggingface.co/datasets/eoguzakin/Robustness of AI-Generated Image Detection Against Localized Inpainting Attacks
 ---
 
 # Robustness of AI-Generated Image Detection Against Localized Inpainting Attacks
@@ -44,6 +44,7 @@ Everything is packaged as `.tar.xz` archives to ensure reproducibility and easy
 │ └─ masks_treering_wm.tar.xz
 └─ checksums.sha256
 
+
 - **detectors/** — per-detector dataset “views,” already resized/re-encoded into the formats expected by each model.  
 - **masks/** — random-rectangle and random-blob object masks (area-binned), used to generate inpainting attacks.  
 - **checksums.sha256** — SHA-256 integrity hashes for all archives.
@@ -71,16 +72,23 @@ randrect/
 randblob_bins/bin{1..4}/
 [reals_inpainted/]
 
-**Detector-specific formats:**
+---
+
+### Detector Input Handling
 
-| Detector | Paradigm | Input format | Notes |
-|----------|----------|--------------|-------|
-| **UFD** (Ojha et al. 2023) | Passive (semantic) | 512×512 PNG | CLIP-based detector |
-| **DIMD** (Corvi et al. 2023) | Passive (artifact) | 256×256 JPEG | Matches training distribution |
-| **DIRE** (Wang et al. 2023) | Training-free (diffusion recon.) | 512×512 PNG | ADM prior |
-| **AEROBLADE** (Ricker et al. 2024) | Training-free (autoencoder recon.) | 512×512 PNG | LPIPS distance |
-| **Stable Signature** (Fernandez et al. 2023) | Watermarking | 512×512 PNG | Watermarked VAE decoder |
-| **Tree-Ring** (Wen et al. 2023) | Watermarking | 512×512 PNG | Frequency-domain watermark |
+**On disk:** All datasets are stored as **PNG, 512×512, lossless**.  
+**At runtime:** Each detector runner applies its own preprocessing to match the original paper/training setup:
+
+- **UFD → 224**  
+  (Resized + center-cropped to 224×224, CLIP normalization.)  
+- **DIMD → JPEG-256**  
+  (Resized to 256×256, with JPEG round-trip to mimic training distribution.)  
+- **DIRE → 256**  
+  (Resized to 256×256, matching the ADM ImageNet-256 diffusion prior.)  
+- **AEROBLADE / StableSig / Tree-Ring → 512**  
+  (All evaluated directly at 512×512 without JPEG compression.)
+
+> **Why this split?** Storing everything as **PNG-512** keeps the dataset lossless and comparable. Each runner then enforces the preprocessing that the corresponding model expects, ensuring scientifically fair evaluation.
 
 ---
 
@@ -143,7 +151,7 @@ Datasets are organized to support a **fixed-threshold robustness evaluation**.
 from huggingface_hub import hf_hub_download
 import tarfile, os
 
-REPO = "eoguzakin/Robustness-of-AI-Generated-Image-Detection-Against-Localized-Inpainting-Attacks"
+REPO = "eoguzakin/Robustness of AI-Generated Image Detection Against Localized Inpainting Attacks"
 
 def fetch_and_extract(filename, target_dir):
     path = hf_hub_download(repo_id=REPO, filename=filename, repo_type="dataset")
@@ -160,57 +168,26 @@ bash
 Kodu kopyala
 sha256sum -c checksums.sha256
 🧪 Provenance
-Reals: SEMI-TRUTHS (Pal et al. 2024) OpenImages subset.
+Reals: SEMI-TRUTHS (Pal et al. 2024), OpenImages subset.
 
 Fakes: GenImage diverse generator set.
 
-Inpainting attacks:
-
-LaMa (Suvorov et al. 2022)
+Inpainting attacks: LaMa (Suvorov et al. 2022), ZITS (Dong et al. 2022).
 
-ZITS (Dong et al. 2022)
+Watermarks: Stable Signature (Fernandez et al. 2023), Tree-Ring (Wen et al. 2023).
 
-Watermarks:
-
-Stable Signature watermarked VAE decoder.
-
-Tree-Ring frequency-domain embedding.
-
-Detector-specific preprocessing (resizing, JPEG re-encoding) applied only in detector views, ensuring comparability.
+Detector-specific preprocessing applied only at runtime, ensuring comparability.
 
 📚 Citations
-If you use this dataset, please cite:
-
-Pal et al., 2024 — Semi-Truths: A Large-Scale Dataset of AI-Augmented Images for Evaluating Robustness of AI-Generated Image Detectors.
-
-Ojha et al., 2023 — Universal Fake Image Detectors.
-
-Corvi et al., 2023 — On the Detection of Synthetic Images Generated by Diffusion Models.
-
-Wang et al., 2023 — DIRE for Diffusion-Generated Image Detection.
-
-Ricker et al., 2024 — AEROBLADE.
-
-Fernandez et al., 2023 — Stable Signature.
-
-Wen et al., 2023 — Tree-Ring Watermarks.
-
-Suvorov et al., 2022 — LaMa Inpainting.
-
-Rombach et al., 2022 — Latent Diffusion Models.
-
-Dong et al., 2022 — ZITS Inpainting.
+(Keep your citation block here as before — unchanged.)
 
 📝 License
-These archives are derived datasets for research use only.
-
+Derived datasets for research use only.
 Upstream datasets (SEMI-TRUTHS, GenImage, LaMa, ZITS, etc.) retain their original licenses.
-
 This packaging (scripts + archive structure) is released under CC BY-NC 4.0 unless otherwise specified.
 
 👤 Maintainer
 Oguz Akin — Saarland University
-
 Contact: [email protected]
 
 🗓️ Changelog