|
--- |
|
datasets: |
|
- ideepankarsharma2003/ImageClassificationStableDiffusion_small |
|
- ideepankarsharma2003/Midjourney_v6_Classification_small_shuffled |
|
- ideepankarsharma2003/AIGeneratedImages_Midjourney |
|
tags: |
|
- image-classification |
|
- ai-gen-images |
|
--- |
|
|
|
|
|
# Model Card for AI Image Classification - Midjourney V6 & SDXL |
|
|
|
## Model Details |
|
|
|
### Model Description |
|
|
|
This model is a **Swin Transformer-based classifier** designed to distinguish between **AI-generated** and **human-created** images, specifically focusing on outputs from **Midjourney V6** and **Stable Diffusion XL (SDXL)**. It has been trained on a curated dataset of AI-generated images. |
|
|
|
- **Developed by:** Deepankar Sharma |
|
- **Model type:** Image Classification (Swin Transformer) |
|
- **Finetuned from model:** SwinForImageClassification |
|
|
|
### Model Sources |
|
|
|
- **Repository:** [Hugging Face Model Repository](https://huggingface.co/ideepankarsharma2003/AI_ImageClassification_MidjourneyV6_SDXL) |
|
|
|
## Uses |
|
|
|
### Direct Use |
|
|
|
This model can be used for **detecting AI-generated images** from Midjourney V6 and SDXL. It is useful for content moderation, fact-checking, and detecting synthetic media. |
|
|
|
### Out-of-Scope Use |
|
|
|
- The model is **not designed** for detecting AI-generated images from all generative models. |
|
- It **may not perform well** on heavily edited AI-generated images or images mixed with human elements. |
|
- It is **not intended for forensic-level deepfake detection**. |
|
|
|
## Bias, Risks, and Limitations |
|
|
|
This model is trained specifically on **Midjourney V6** and **Stable Diffusion XL** datasets. It may not generalize well to images generated by other AI models. Additionally, biases in the dataset could lead to **false positives** (flagging real images as AI-generated) or **false negatives** (failing to detect AI-generated content). |
|
|
|
### Recommendations |
|
|
|
Users should verify results with additional tools and **not solely rely on this model** for high-stakes decisions. Model performance should be tested on domain-specific datasets before deployment. |
|
|
|
## How to Get Started with the Model |
|
|
|
You can use this model with the 🤗 Transformers library: |
|
|
|
```python |
|
from transformers import AutoModelForImageClassification, AutoFeatureExtractor |
|
from PIL import Image |
|
import torch |
|
|
|
# Load model and feature extractor |
|
model_name = "ideepankarsharma2003/AI_ImageClassification_MidjourneyV6_SDXL" |
|
model = AutoModelForImageClassification.from_pretrained(model_name) |
|
feature_extractor = AutoFeatureExtractor.from_pretrained(model_name) |
|
|
|
# Load and preprocess image |
|
image = Image.open("path_to_image.jpg") |
|
inputs = feature_extractor(images=image, return_tensors="pt") |
|
|
|
# Perform inference |
|
with torch.no_grad(): |
|
outputs = model(**inputs) |
|
logits = outputs.logits |
|
predicted_label = logits.argmax(-1).item() |
|
|
|
# Label Mapping |
|
id2label = {0: "ai_gen", 1: "human"} |
|
print("Predicted label:", id2label[predicted_label]) |
|
``` |
|
|
|
## Training Details |
|
|
|
### Training Data |
|
|
|
The model was trained on the following datasets: |
|
|
|
- [ImageClassificationStableDiffusion_small](https://huggingface.co/datasets/ideepankarsharma2003/ImageClassificationStableDiffusion_small) |
|
- [Midjourney_v6_Classification_small_shuffled](https://huggingface.co/datasets/ideepankarsharma2003/Midjourney_v6_Classification_small_shuffled) |
|
- [AIGeneratedImages_Midjourney](https://huggingface.co/datasets/ideepankarsharma2003/AIGeneratedImages_Midjourney) |
|
|
|
### Training Procedure |
|
|
|
- **Image Size:** 224x224 |
|
- **Patch Size:** 4 |
|
- **Embedding Dimension:** 128 |
|
- **Layers:** 4 |
|
- **Attention Heads per Stage:** [4, 8, 16, 32] |
|
- **Dropout Rates:** |
|
- Attention: 0.0 |
|
- Hidden: 0.0 |
|
- Drop Path: 0.1 |
|
- **Activation Function:** GeLU |
|
- **Optimizer:** AdamW |
|
- **Learning Rate Scheduler:** Cosine Annealing |
|
- **Precision:** float32 |
|
- **Training Steps:** 3414 |
|
|
|
## Evaluation |
|
|
|
### Testing Data, Factors & Metrics |
|
|
|
#### Testing Data |
|
|
|
The model was evaluated on a separate validation split from the training datasets. |
|
|
|
#### Metrics |
|
|
|
- **Accuracy** |
|
- **Precision & Recall** |
|
- **F1 Score** |
|
|
|
### Summary |
|
|
|
The model effectively distinguishes between AI-generated and human-created images, but its performance may be affected by dataset biases and out-of-distribution examples. |
|
|
|
## Citation |
|
|
|
If you use this model, please cite: |
|
|
|
```bibtex |
|
@misc{ai_image_classification, |
|
author = {Deepankar Sharma}, |
|
title = {AI Image Classification - Midjourney V6 & SDXL}, |
|
year = {2024}, |
|
publisher = {Hugging Face}, |
|
howpublished = {\url{https://huggingface.co/ideepankarsharma2003/AI_ImageClassification_MidjourneyV6_SDXL}} |
|
} |
|
``` |
|
|
|
## Model Card Authors |
|
|
|
- **Author:** Deepankar Sharma |
|
|
|
--- |
