some changes
Browse files- app.py +8 -15
- cheetah.jpg +0 -0
- horse.jpg +0 -0
app.py
CHANGED
|
@@ -5,12 +5,12 @@ import torchvision.models as models
|
|
| 5 |
from torchvision.transforms import v2 as transforms
|
| 6 |
import os
|
| 7 |
|
|
|
|
| 8 |
class_names = ['AI-Generated Image', "Real/Non-AI-Generated Image"]
|
| 9 |
|
| 10 |
-
#
|
| 11 |
-
# model = models.vit_b_16()
|
| 12 |
weights_path = "FaKe-ViT-B16.pth"
|
| 13 |
-
model = torch.load(weights_path
|
| 14 |
model.eval()
|
| 15 |
# Preprocessing the image
|
| 16 |
preprocess = transforms.Compose([
|
|
@@ -22,7 +22,6 @@ preprocess = transforms.Compose([
|
|
| 22 |
|
| 23 |
# Define the prediction function
|
| 24 |
def predict_image(image):
|
| 25 |
-
# inp = Image.fromarray(inp.astype('uint8'), 'RGB')
|
| 26 |
image = preprocess(image)
|
| 27 |
if image.shape[0] != 3:
|
| 28 |
image = image[:3, :, :]
|
|
@@ -32,8 +31,6 @@ def predict_image(image):
|
|
| 32 |
output1 = torch.argmax(torch.softmax(output,dim=1),dim=1).item()
|
| 33 |
return class_names[output1]
|
| 34 |
|
| 35 |
-
# def image_mod(image):
|
| 36 |
-
# return image.rotate(45)
|
| 37 |
|
| 38 |
|
| 39 |
demo = gr.Interface(
|
|
@@ -41,17 +38,13 @@ demo = gr.Interface(
|
|
| 41 |
gr.Image(image_mode="RGB",type="pil"),
|
| 42 |
"text",
|
| 43 |
flagging_options=["incorrect prediction"],
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
# ],
|
| 50 |
title="FaKe-ViT-B/16: AI-Generated Image Detection using Vision Transformer(ViT-B/16)",
|
| 51 |
description="This is a demo to detect AI-Generated images using Vision Transformer(ViT-B/16). Upload an image and the model will predict whether the image is AI-Generated or Real",
|
| 52 |
-
css=""".gr-header, .gr-text {
|
| 53 |
-
font-size: 20px;
|
| 54 |
-
}""",
|
| 55 |
article=" \nBased on the paper:'An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale', Alexey et al.\nDataset: 'Fake or Real competition dataset' at https://huggingface.co/datasets/mncai/Fake_or_Real_Competition_Dataset"
|
| 56 |
)
|
| 57 |
|
|
|
|
| 5 |
from torchvision.transforms import v2 as transforms
|
| 6 |
import os
|
| 7 |
|
| 8 |
+
# Define the class names
|
| 9 |
class_names = ['AI-Generated Image', "Real/Non-AI-Generated Image"]
|
| 10 |
|
| 11 |
+
# Load the model
|
|
|
|
| 12 |
weights_path = "FaKe-ViT-B16.pth"
|
| 13 |
+
model = torch.load(weights_path, map_location=torch.device('cpu'))
|
| 14 |
model.eval()
|
| 15 |
# Preprocessing the image
|
| 16 |
preprocess = transforms.Compose([
|
|
|
|
| 22 |
|
| 23 |
# Define the prediction function
|
| 24 |
def predict_image(image):
|
|
|
|
| 25 |
image = preprocess(image)
|
| 26 |
if image.shape[0] != 3:
|
| 27 |
image = image[:3, :, :]
|
|
|
|
| 31 |
output1 = torch.argmax(torch.softmax(output,dim=1),dim=1).item()
|
| 32 |
return class_names[output1]
|
| 33 |
|
|
|
|
|
|
|
| 34 |
|
| 35 |
|
| 36 |
demo = gr.Interface(
|
|
|
|
| 38 |
gr.Image(image_mode="RGB",type="pil"),
|
| 39 |
"text",
|
| 40 |
flagging_options=["incorrect prediction"],
|
| 41 |
+
examples=[
|
| 42 |
+
os.path.join(os.path.dirname(__file__), "images/cheetah.jpg"),
|
| 43 |
+
os.path.join(os.path.dirname(__file__), "images/horse.jpg"),
|
| 44 |
+
os.path.join(os.path.dirname(__file__), "images/astronaut.png"),
|
| 45 |
+
],
|
|
|
|
| 46 |
title="FaKe-ViT-B/16: AI-Generated Image Detection using Vision Transformer(ViT-B/16)",
|
| 47 |
description="This is a demo to detect AI-Generated images using Vision Transformer(ViT-B/16). Upload an image and the model will predict whether the image is AI-Generated or Real",
|
|
|
|
|
|
|
|
|
|
| 48 |
article=" \nBased on the paper:'An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale', Alexey et al.\nDataset: 'Fake or Real competition dataset' at https://huggingface.co/datasets/mncai/Fake_or_Real_Competition_Dataset"
|
| 49 |
)
|
| 50 |
|
cheetah.jpg
ADDED
|
horse.jpg
ADDED
|