update

README.md

@@ -10,43 +10,55 @@ pinned: false
 license: mit
 ---
 
-# Deepfake Detection Space
+# Deepfake Detection Space (Open-Source Demonstrator)
 
-This Space provides a unified interface to test multiple state-of-the-art deepfake detection models on your images. These are the detectors that you can find in the [Image-Deepfake-Detectors-Public-Library](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
+Welcome to the Truebees Deepfake Detection Space! This Space provides a unified, user-friendly interface to test multiple open-source state-of-the-art deepfake detection models on your own images. 
 
-## Available Detectors
+These detectors are available in the [Image-Deepfake-Detectors-Public-Library](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
 
-- **R50_TF** - ResNet-50 based detector trained on TrueFake dataset
-- **R50_nodown** - ResNet-50 without downsampling operations
-- **CLIP-D** - CLIP-based deepfake detector
-- **P2G** - Prompt2Guard: Conditioned prompt-optimization for continual deepfake detection
-- **NPR** - Neural Posterior Regularization
+## 🎯 Goal and Context
 
-## Usage
+This demonstrator is designed as an **educational and open-source contribution** to the community. We want to allow users, researchers, and developers to experiment with baseline forensic models for free, familiarize themselves with AI detection tools, and understand both their capabilities and their limitations.
 
-1. Upload an image
-2. Select a detector from the dropdown
-3. Click "Detect" to get the prediction
+**⚠️ Important Limitations & Disclaimer:**
+* **Non-Social Training Data:** The models included here were trained specifically on non-social datasets. 
+* **"In-the-Wild" Performance Drop:** Because these models lack training on highly compressed data, you may experience a severe drop in performance or inaccurate results when testing images downloaded from social media platforms or messaging apps (e.g., WhatsApp). 
+* **Not Proprietary Software:** This Space showcases open-source baseline networks. It **does not** reflect the performance, robustness, or accuracy of Truebees' proprietary commercial detection methods, which are designed to handle complex "in-the-wild" manipulations.
 
-The detector will return:
-- **Prediction**: Real or Fake
-- **Confidence**: Model confidence score (0-1)
-- **Elapsed Time**: Processing time
+## 🛠️ Available Detectors
 
-## Models
+* **R50_TF**: ResNet-50 based detector trained on the TrueFake dataset.
+* **R50_nodown**: ResNet-50 without downsampling operations.
+* **CLIP-D**: CLIP-based deepfake detector.
+* **P2G**: Prompt2Guard: Conditioned prompt-optimization for continual deepfake detection.
+* **NPR**: Neural Posterior Regularization.
 
-All models have been pretrained on images generated with StyleGAN2 and StableDiffusionXL, and real images from the FFHQ Dataset and the FORLAB Dataset.
+## 🚀 Usage Guide
 
-## References
+1. **Upload an image**: Choose an image you want to analyze.
+2. **Select a detector**: Choose one of the open-source models from the dropdown menu.
+3. **Detect**: Click the "Detect" button to generate the prediction.
 
-For more information about the implementation and benchmarking, visit the [GitHub repository](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
+**Understanding the Output:**
+* **Prediction**: Indicates whether the model believes the image is "Real" or "Fake".
+* **Confidence**: The model's confidence score, ranging from 0 to 1.
+* **Elapsed Time**: Total processing time taken by the model.
 
-## Acknowledgment
+## 🧠 Training Data Overview
 
-This work has been funded by the European Union under Horizon Europe Grant Agreement no. 101092887 (NGI Sargasso – Open Call OC5 | Deepshield Project).
-Website page: [NGI Sargasso Innovators](https://ngisargasso.eu/2025/02/28/discover-the-last-13-ngi-sargasso-innovators-from-our-open-call-5-creating-an-internet-of-humans-across-continents/)
+All open-source models featured in this Space have been pretrained on specific datasets:
+* **Fake Images:** Generated using StyleGAN2 and StableDiffusionXL.
+* **Real Images:** Sourced from the FFHQ Dataset and the FORLAB Dataset.
 
-## Note
+## 📚 References
 
-⚠️ Due to file size limitations, model weights need to be downloaded automatically on first use. This may take a few moments.
+For technical details, implementation specifics, and benchmarking data, please visit our [GitHub repository](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
 
+## 🤝 Acknowledgment
+
+This work has been funded by the European Union under Horizon Europe Grant Agreement no. 101092887 (NGI Sargasso – Open Call OC5 | Deepshield Project). 
+* **Website page**: [NGI Sargasso Innovators](https://ngisargasso.eu/2025/02/28/discover-the-last-13-ngi-sargasso-innovators-from-our-open-call-5-creating-an-internet-of-humans-across-continents/)
+
+## ⏳ Note
+
+⚠️ Due to file size limitations, model weights need to be downloaded automatically upon first use. This may take a few moments.

README_HF.md

@@ -1,8 +1,8 @@
 ---
-title: Deepfake Detection Library
+title: Deepfake Detection Space
 emoji: 🔍
 colorFrom: red
-colorTo: orange
+colorTo: yellow
 sdk: gradio
 sdk_version: 5.49.1
 app_file: app.py
@@ -10,37 +10,55 @@ pinned: false
 license: mit
 ---
 
-# Deepfake Detection Library
+# Deepfake Detection Space (Open-Source Demonstrator)
 
-This Space provides a unified interface to test multiple state-of-the-art deepfake detection models on your images.
+Welcome to the Truebees Deepfake Detection Space! This Space provides a unified, user-friendly interface to test multiple open-source state-of-the-art deepfake detection models on your own images. 
 
-## Available Detectors
+These detectors are available in the [Image-Deepfake-Detectors-Public-Library](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
 
-- **R50_TF** - ResNet-50 based detector trained on TrueFake dataset
-- **R50_nodown** - ResNet-50 without downsampling operations
-- **CLIP-D** - CLIP-based deepfake detector
-- **P2G** - Prompt2Guard: Conditioned prompt-optimization for continual deepfake detection
-- **NPR** - Neural Posterior Regularization
+## 🎯 Goal and Context
 
-## Usage
+This demonstrator is designed as an **educational and open-source contribution** to the community. We want to allow users, researchers, and developers to experiment with baseline forensic models for free, familiarize themselves with AI detection tools, and understand both their capabilities and their limitations.
 
-1. Upload an image
-2. Select a detector from the dropdown
-3. Click "Detect" to get the prediction
+**⚠️ Important Limitations & Disclaimer:**
+* **Non-Social Training Data:** The models included here were trained specifically on non-social datasets. 
+* **"In-the-Wild" Performance Drop:** Because these models lack training on highly compressed data, you may experience a severe drop in performance or inaccurate results when testing images downloaded from social media platforms or messaging apps (e.g., WhatsApp). 
+* **Not Proprietary Software:** This Space showcases open-source baseline networks. It **does not** reflect the performance, robustness, or accuracy of Truebees' proprietary commercial detection methods, which are designed to handle complex "in-the-wild" manipulations.
 
-The detector will return:
-- **Prediction**: Real or Fake
-- **Confidence**: Model confidence score (0-1)
-- **Elapsed Time**: Processing time
+## 🛠️ Available Detectors
 
-## Models
+* **R50_TF**: ResNet-50 based detector trained on the TrueFake dataset.
+* **R50_nodown**: ResNet-50 without downsampling operations.
+* **CLIP-D**: CLIP-based deepfake detector.
+* **P2G**: Prompt2Guard: Conditioned prompt-optimization for continual deepfake detection.
+* **NPR**: Neural Posterior Regularization.
 
-All models have been pretrained on images generated with StyleGAN2 and StableDiffusionXL, and real images from the FFHQ Dataset and the FORLAB Dataset.
+## 🚀 Usage Guide
 
-## References
+1. **Upload an image**: Choose an image you want to analyze.
+2. **Select a detector**: Choose one of the open-source models from the dropdown menu.
+3. **Detect**: Click the "Detect" button to generate the prediction.
 
-For more information about the implementation and benchmarking, visit the [GitHub repository](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
+**Understanding the Output:**
+* **Prediction**: Indicates whether the model believes the image is "Real" or "Fake".
+* **Confidence**: The model's confidence score, ranging from 0 to 1.
+* **Elapsed Time**: Total processing time taken by the model.
 
-## Note
+## 🧠 Training Data Overview
 
-⚠️ Due to file size limitations, model weights need to be downloaded automatically on first use. This may take a few moments.
+All open-source models featured in this Space have been pretrained on specific datasets:
+* **Fake Images:** Generated using StyleGAN2 and StableDiffusionXL.
+* **Real Images:** Sourced from the FFHQ Dataset and the FORLAB Dataset.
+
+## 📚 References
+
+For technical details, implementation specifics, and benchmarking data, please visit our [GitHub repository](https://github.com/truebees-ai/Image-Deepfake-Detectors-Public-Library).
+
+## 🤝 Acknowledgment
+
+This work has been funded by the European Union under Horizon Europe Grant Agreement no. 101092887 (NGI Sargasso – Open Call OC5 | Deepshield Project). 
+* **Website page**: [NGI Sargasso Innovators](https://ngisargasso.eu/2025/02/28/discover-the-last-13-ngi-sargasso-innovators-from-our-open-call-5-creating-an-internet-of-humans-across-continents/)
+
+## ⏳ Note
+
+⚠️ Due to file size limitations, model weights need to be downloaded automatically upon first use. This may take a few moments.

app.py

@@ -268,26 +268,29 @@ def predict(image_path, detector_name):
 demo = gr.Blocks(title="Deepfake Detection Space", theme=gr.themes.Soft())
 
 with demo:
-    gr.Markdown("# 🔍 Deepfake Detection Space")
     gr.Markdown("""
-    This space collects a series of state-of-the-art methods for deepfake detection, allowing for free and unlimited use.
+    # 🔍 Deepfake Detection Space
+    This space provides an **educational, open-source demonstrator** collecting a series of baseline state-of-the-art methods for deepfake detection, allowing for free and unlimited use.
     
+    ### ⚠️ Important Limitations
+    * **Not Proprietary Software:** This tool showcases open-source networks. It **does not** reflect the performance, robustness, or accuracy of Truebees' commercial detection methods.
+    * **"In-the-Wild" Performance Drop:** These models were trained on non-social datasets. You may experience a severe drop in accuracy when testing highly compressed images downloaded from social media platforms or messaging apps (e.g., WhatsApp).
+
     ### Training & Performance
     All methods have been trained using the **[DeepShield dataset](https://zenodo.org/records/15648378)**, on images generated with **Stable Diffusion XL** and **StyleGAN 2**. 
     You can expect performance comparable to the results shown in [Dell'Anna et al. (2025)](https://arxiv.org/pdf/2504.20658).
 
     ### Understanding the Results
-    *   **Prediction**: Tells if an image is **Real** or **Fake**.
-    *   **Confidence**: The confidence with which the model determines if the image is real or fake.
-    *   **Elapsed Time**: The time the model needed to make the prediction (excluding preprocessing or model building).
+    * **Prediction**: Tells if an image is **Real** or **Fake**.
+    * **Confidence**: The confidence with which the model determines if the image is real or fake.
+    * **Elapsed Time**: The time the model needed to make the prediction (excluding preprocessing or model building).
 
     ### Understanding the Results produced by "ALL"
-    *   Runs all available detectors (R50_TF, R50_nodown, CLIP-D, P2G, NPR) sequentially on the input image.
-    *   Produces a **Weighted Majority Vote** verdict (Real/Fake). Each model's vote is weighted by a fixed importance score (summing to 1) based on user ranking **and its confidence score**. Only confident predictions (> 0.6) are counted.
-    *   You can find the specific weights used for each model in the **"⚖️ Weight Details"** menu below.
-    *   Also generates a **Confidence Plot** visualizing each model's score and a textual **Explanation** of the consensus.
-    *   In the plot, **Green** bars indicate a **Real** prediction, while **Red** bars indicate a **Fake** prediction.
-
+    * Runs all available detectors (R50_TF, R50_nodown, CLIP-D, P2G, NPR) sequentially on the input image.
+    * Produces a **Weighted Majority Vote** verdict (Real/Fake). Each model's vote is weighted by a fixed importance score (summing to 1) based on user ranking **and its confidence score**. Only confident predictions (> 0.6) are counted.
+    * You can find the specific weights used for each model in the **"⚖️ Weight Details"** menu below.
+    * Also generates a **Confidence Plot** visualizing each model's score and a textual **Explanation** of the consensus.
+    * In the plot, **Green** bars indicate a **Real** prediction, while **Red** bars indicate a **Fake** prediction.
 
     ### Note
     ⚠️ Due to file size limitations, model weights need to be downloaded automatically on first use. This may take a few moments. <br>
@@ -331,33 +334,33 @@ with demo:
     with gr.Accordion("📚 Model Details", open=False):
         gr.Markdown("""
         ### **ALL**
-        *   **Description**: Runs all available detectors (R50_TF, R50_nodown, CLIP-D, P2G, NPR) sequentially on the input image.
-        *   **Results**: Produces a **Majority Vote** verdict (Real/Fake) considering only confident predictions (> 0.6). Also generates a **Confidence Plot** visualizing each model's score and a textual **Explanation** of the consensus.
+        * **Description**: Runs all available detectors (R50_TF, R50_nodown, CLIP-D, P2G, NPR) sequentially on the input image.
+        * **Results**: Produces a **Majority Vote** verdict (Real/Fake) considering only confident predictions (> 0.6). Also generates a **Confidence Plot** visualizing each model's score and a textual **Explanation** of the consensus.
 
         ### **R50_TF**
-        *   **Description**: A ResNet50 architecture modified to exclude downsampling at the first layer. It uses "learned prototypes" in the classification head for robust detection.
-        *   **Paper**: [TrueFake: A Real World Case Dataset of Last Generation Fake Images also Shared on Social Networks](https://arxiv.org/pdf/2504.20658)
-        *   **Code**: [GitHub Repository](https://github.com/MMLab-unitn/TrueFake-IJCNN25)
+        * **Description**: A ResNet50 architecture modified to exclude downsampling at the first layer. It uses "learned prototypes" in the classification head for robust detection.
+        * **Paper**: [TrueFake: A Real World Case Dataset of Last Generation Fake Images also Shared on Social Networks](https://arxiv.org/pdf/2504.20658)
+        * **Code**: [GitHub Repository](https://github.com/MMLab-unitn/TrueFake-IJCNN25)
 
         ### **R50_nodown**
-        *   **Description**: A ResNet-50 model without downsampling operations in the first layer, designed to preserve high-frequency artifacts common in synthetic images.
-        *   **Paper**: [On the detection of synthetic images generated by diffusion models](https://arxiv.org/abs/2211.00680)
-        *   **Code**: [GitHub Repository](https://grip-unina.github.io/DMimageDetection/)
+        * **Description**: A ResNet-50 model without downsampling operations in the first layer, designed to preserve high-frequency artifacts common in synthetic images.
+        * **Paper**: [On the detection of synthetic images generated by diffusion models](https://arxiv.org/abs/2211.00680)
+        * **Code**: [GitHub Repository](https://grip-unina.github.io/DMimageDetection/)
 
         ### **CLIP-D**
-        *   **Description**: A lightweight detection strategy based on CLIP features. It exhibits surprising generalization ability using only a handful of example images.
-        *   **Paper**: [Raising the Bar of AI-generated Image Detection with CLIP](https://arxiv.org/abs/2312.00195v2)
-        *   **Code**: [GitHub Repository](https://grip-unina.github.io/ClipBased-SyntheticImageDetection/)
+        * **Description**: A lightweight detection strategy based on CLIP features. It exhibits surprising generalization ability using only a handful of example images.
+        * **Paper**: [Raising the Bar of AI-generated Image Detection with CLIP](https://arxiv.org/abs/2312.00195v2)
+        * **Code**: [GitHub Repository](https://grip-unina.github.io/ClipBased-SyntheticImageDetection/)
 
         ### **P2G (Prompt2Guard)**
-        *   **Description**: Uses Vision-Language Models (VLMs) with conditioned prompt-optimization for continual deepfake detection. It leverages read-only prompts for efficiency.
-        *   **Paper**: [Conditioned Prompt-Optimization for Continual Deepfake Detection](https://arxiv.org/abs/2407.21554)
-        *   **Code**: [GitHub Repository](https://github.com/laitifranz/Prompt2Guard)
+        * **Description**: Uses Vision-Language Models (VLMs) with conditioned prompt-optimization for continual deepfake detection. It leverages read-only prompts for efficiency.
+        * **Paper**: [Conditioned Prompt-Optimization for Continual Deepfake Detection](https://arxiv.org/abs/2407.21554)
+        * **Code**: [GitHub Repository](https://github.com/laitifranz/Prompt2Guard)
 
         ### **NPR**
-        *   **Description**: Focuses on Neighboring Pixel Relationships (NPR) to capture generalized structural artifacts stemming from up-sampling operations in generative networks.
-        *   **Paper**: [Rethinking the Up-Sampling Operations in CNN-based Generative Network for Generalizable Deepfake Detection](https://arxiv.org/abs/2312.10461)
-        *   **Code**: [GitHub Repository](https://github.com/chuangchuangtan/NPR-DeepfakeDetection)
+        * **Description**: Focuses on Neighboring Pixel Relationships (NPR) to capture generalized structural artifacts stemming from up-sampling operations in generative networks.
+        * **Paper**: [Rethinking the Up-Sampling Operations in CNN-based Generative Network for Generalizable Deepfake Detection](https://arxiv.org/abs/2312.10461)
+        * **Code**: [GitHub Repository](https://github.com/chuangchuangtan/NPR-DeepfakeDetection)
         """)
 
     gr.Markdown("""
@@ -382,4 +385,4 @@ if __name__ == "__main__":
         demo.launch(server_name="0.0.0.0", server_port=7860, allowed_paths=["."])
     else:
         # Local execution
-        demo.launch()
+        demo.launch()