docs: overhaul README — fix notice, ONNX variant guide, test app, deprecate old ONNX repo

LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) 2024 Jeongsoo Park, Andrew Owens, University of Michigan
+Copyright (c) 2025 Borderless (HuggingFace integration, ONNX exports, configuration)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -14,78 +14,154 @@ tags:
 - deepfake_detection
 - community
 - opensight
+- onnx
 ---
 
-# NOTICE: 7/22/2026
+# CommunityForensics DeepfakeDet-ViT
 
-- Ungated again. Friendly reminder to all to please respect the MIT license. Attribution should be made to the original project (see bottom) as well as this repository. Other than that, you're free to use the model as you wish.
+Vision Transformer (ViT-Small) trained on 2.7M samples across 4,803 generators for detecting AI-generated images. Presented in [Community Forensics: Using Thousands of Generators to Train Fake Image Detectors](https://huggingface.co/papers/2411.04125) (CVPR 2025).
 
-## Trained on 2.7M samples across 4,803 generators (see Training Data)
+## IMPORTANT — Configuration Fix (July 2026)
 
-Model presented in [Community Forensics: Using Thousands of Generators to Train Fake Image Detectors](https://huggingface.co/papers/2411.04125).
+**If you downloaded this model before July 22, 2026, your local copy has incorrect config files.** The model weights were always correct, but the metadata was wrong. This silently degraded results or caused loading errors.
 
-**Uploaded for community validation as part of OpenSight** - An upcoming open-source framework for adaptive deepfake detection.
+| Bug | Effect | Fixed Value |
+|---|---|---|
+| `num_classes: 1` | `from_pretrained()` **crashes** — classifier weight is `[2,384]`, not `[1,384]` | `2` |
+| `num_attention_heads: 12` | **Silently wrong** — attention sliced 12×32d instead of correct 6×64d | `6` |
+| Preprocessor `size: 384` | No center-crop — images processed differently than training | `{height:440, width:440}` + `do_center_crop` |
+| Missing `id2label` / `label2id` | Raw `[0,1]` output unreadable | `{"0":"real","1":"fake"}` |
 
-**Project OpenSight HF Spaces coming soon with an eval playground and eventually a leaderboard. Preview:** 
+### How to verify you have the fix
 
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/639daf827270667011153fbc/AUmW697OefKN83BClM1ae.png)
+```python
+import json
+with open("path/to/config.json") as f:
+    cfg = json.load(f)
+assert cfg["num_classes"] == 2, "Still broken — re-download the model"
+assert cfg["num_attention_heads"] == 6, "Still broken — re-download the model"
+```
 
-## Model Details
-### Model Description
-Vision Transformer (ViT) model trained on the largest dataset to-date for detecting AI-generated images in forensic applications.
+### One-liner migration
+
+```bash
+# Re-download the repo to get the fixed configs
+git clone https://huggingface.co/buildborderless/CommunityForensics-DeepfakeDet-ViT
+# Or if you use huggingface_hub:
+# huggingface-cli download buildborderless/CommunityForensics-DeepfakeDet-ViT --local-dir ./
+```
+
+### If you were using the old custom wrapper (`modeling_vit_classifier.py`)
+
+It has been moved to `scripts/` and marked deprecated. Switch to the standard HuggingFace path:
+
+```python
+from transformers import ViTForImageClassification, ViTImageProcessor
+model = ViTForImageClassification.from_pretrained("buildborderless/CommunityForensics-DeepfakeDet-ViT")
+processor = ViTImageProcessor.from_pretrained("buildborderless/CommunityForensics-DeepfakeDet-ViT")
+```
+
+### If you were using the ONNX repo
+
+The separate [`buildborderless/CommunityForensics-DeepfakeDet-ViT-ONNX`](https://huggingface.co/buildborderless/CommunityForensics-DeepfakeDet-ViT-ONNX) repo is now deprecated. All ONNX models (including quantized variants) are now included here in the `onnx/` directory with corrected configs.
+
+---
+
+## Quick Start
+
+```python
+from transformers import ViTForImageClassification, ViTImageProcessor
+from PIL import Image
+
+model = ViTForImageClassification.from_pretrained("buildborderless/CommunityForensics-DeepfakeDet-ViT")
+processor = ViTImageProcessor.from_pretrained("buildborderless/CommunityForensics-DeepfakeDet-ViT")
+
+image = Image.open("suspicious_image.jpg")
+inputs = processor(image, return_tensors="pt")
+outputs = model(**inputs)
+
+import torch
+probs = torch.softmax(outputs.logits, dim=-1)[0]
+print(f"real: {probs[0]:.4f}, fake: {probs[1]:.4f}")
+print(f"verdict: {model.config.id2label[torch.argmax(probs).item()]}")
+```
 
-- **Developed by:** Jeongsoo Park and Andrew Owens, University of Michigan
-- **Model type:** Vision Transformer (ViT-Small)
-- **License:** MIT (compatible with CreativeML OpenRAIL-M referenced in [2411.04125v1.pdf])
-- **Finetuned from:** timm/vit_small_patch16_384.augreg_in21k_ft_in1k
-- **Adapted for HF** inference compatibility by Borderless.
+---
+
+## ONNX Variants
+
+Eight pre-exported ONNX models with different size/speed/accuracy trade-offs. All use the corrected 2-class config.
+
+| Variant | Size | Speed (CPU) | Accuracy | Best For |
+|---|---|---|---|---|
+| `model.onnx` (full) | 138 MB | ★★★ | ★★★★★ | Server-side, maximum accuracy |
+| `model_fp16.onnx` | 69 MB | ★★ | ★★★★★ | GPU inference, near-lossless |
+| `model_int8.onnx` | 36 MB | ★★★★★ | ★★★★ | Fastest CPU, balanced |
+| `model_uint8.onnx` | 36 MB | ★★★★★ | ★★★★ | Fast CPU, unsigned integer |
+| `model_quantized.onnx` | 36 MB | ★★★★ | ★★★★ | General CPU deployment |
+| `model_q4.onnx` | 24 MB | ★★★ | ★★★ | Low-memory, decent accuracy |
+| `model_bnb4.onnx` | 22 MB | ★★ | ★★★ | Constrained edge devices |
+| `model_q4f16.onnx` | 21 MB | ★★ | ★★★ | Smallest file, mobile/web |
+
+**Key insight**: On CPU, INT8 variants are fastest (optimized kernels). Smaller 4-bit models are slower due to dequantization overhead — use them only when disk/RAM is the bottleneck.
+
+```python
+import onnxruntime as ort
+import numpy as np
+
+session = ort.InferenceSession("onnx/model_int8.onnx")
+# Preprocess image: resize to 440, center-crop to 384, normalize
+image = Image.open("image.jpg").resize((440, 440))
+# ... center-crop + normalize ...
+output = session.run(None, {"pixel_values": input_array})
+logits = output[0][0]  # shape [2]: [real_score, fake_score]
+```
 
-**HF Space will be open sourced shortly showcasing various ways to run ultra-fast inference. Make sure to follow us for updates, as we will be releasing a slew of projects in the coming weeks.**
+---
+
+## Model Details
+
+- **Developed by**: Jeongsoo Park and Andrew Owens, University of Michigan
+- **HF integration + ONNX**: Borderless
+- **Model type**: Vision Transformer (ViT-Small)
+- **License**: MIT
+- **Input**: RGB image, resized to 440×440, center-cropped to 384×384, CLIP-normalized
+- **Output**: 2-class logits `[real, fake]`
+- **Architecture**: hidden_size=384, 6 attention heads, 12 layers, patch_size=16
 
 ### Links
-- **Repository:** [JeongsooP/Community-Forensics](https://github.com/JeongsooP/Community-Forensics)
-- **Paper:** [arXiv:2411.04125](https://arxiv.org/pdf/2411.04125)
-- **Project Page:** https://jespark.net/projects/2024/community_forensics
 
-## Training Details
-### Training Data
-- 2.7mil images from 15+ generators, 4600+ models
-- Over 1.15TB worth of images
+- **Original paper**: [arXiv:2411.04125](https://arxiv.org/pdf/2411.04125)
+- **Original repository**: [JeongsooP/Community-Forensics](https://github.com/JeongsooP/Community-Forensics)
+- **Project page**: https://jespark.net/projects/2024/community_forensics
+- **Datasets**: [Full (1.1TB)](https://huggingface.co/datasets/OwensLab/CommunityForensics), [Small (278GB)](https://huggingface.co/datasets/OwensLab/CommunityForensics-Small), [Eval (206GB)](https://huggingface.co/datasets/OwensLab/CommunityForensics-Eval)
 
-### Training Hyperparameters
-- **Framework:** PyTorch 2.0
-- **Precision:** bf16 mixed
-- **Optimizer:** AdamW (lr=5e-5)
-- **Epochs:** 10
-- **Batch Size:** 32
+---
 
-## Evaluation
-### Unverified Testing Results
-- Only unverified because we currently lack resources to evaluate a dataset over 1.4T large.
+## Local Test App
 
-| Metric        | Value |
-|---------------|-------|
-| Accuracy      | 97.2% |
-| F1 Score      | 0.968 |
-| AUC-ROC       | 0.992 |
-| FP Rate       | 2.1%  |
+A local Gradio app for testing PyTorch and ONNX models side by side with benchmarks:
 
-![image/png](https://cdn-uploads.huggingface.co/production/uploads/639daf827270667011153fbc/g-dLzxLBw1RAuiplvFCxh.png)
+```bash
+git clone https://huggingface.co/buildborderless/CommunityForensics-DeepfakeDet-ViT
+cd CommunityForensics-DeepfakeDet-ViT
+pip install -r test_app/requirements.txt
+python test_app/app.py
+```
 
-## Re-sampled and refined dataset
+Three tabs: PyTorch vs ONNX comparison, Benchmark (all 8 ONNX variants on test images), and help.
 
-- **Coming soon™**
+---
 
 ## Citation
-**BibTeX:**
+
 ```bibtex
-@misc{park2024communityforensics,
-    title={Community Forensics: Using Thousands of Generators to Train Fake Image Detectors}, 
-    author={Jeongsoo Park and Andrew Owens},
-    year={2024},
-    eprint={2411.04125},
-    archivePrefix={arXiv},
-    primaryClass={cs.CV},
-    url={https://arxiv.org/abs/2411.04125}, 
+@InProceedings{Park_2025_CVPR,
+    author    = {Park, Jeongsoo and Owens, Andrew},
+    title     = {Community Forensics: Using Thousands of Generators to Train Fake Image Detectors},
+    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
+    month     = {June},
+    year      = {2025},
+    pages     = {8245-8257}
 }
-```
+```