Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -5,6 +5,7 @@ import torchvision.models as models
|
|
5 |
import torchvision.transforms as transforms
|
6 |
from PIL import Image
|
7 |
from huggingface_hub import hf_hub_download
|
|
|
8 |
|
9 |
########################################
|
10 |
# 1. Define the Model Architecture
|
@@ -13,9 +14,7 @@ class MultiTaskModel(nn.Module):
|
|
13 |
def __init__(self, backbone, feature_dim, num_obj_classes):
|
14 |
super(MultiTaskModel, self).__init__()
|
15 |
self.backbone = backbone
|
16 |
-
# Object recognition head
|
17 |
self.obj_head = nn.Linear(feature_dim, num_obj_classes)
|
18 |
-
# Binary classification head (0: AI-generated, 1: Real)
|
19 |
self.bin_head = nn.Linear(feature_dim, 2)
|
20 |
|
21 |
def forward(self, x):
|
@@ -27,41 +26,33 @@ class MultiTaskModel(nn.Module):
|
|
27 |
########################################
|
28 |
# 2. Reconstruct the Model and Load Weights
|
29 |
########################################
|
30 |
-
|
31 |
-
num_obj_classes = 494 # Updated to match the state dict from training
|
32 |
-
|
33 |
device = torch.device("cpu")
|
34 |
|
35 |
-
# Instantiate the backbone: a ResNet-50 with its final layer removed.
|
36 |
resnet = models.resnet50(pretrained=False)
|
37 |
-
resnet.fc = nn.Identity()
|
38 |
feature_dim = 2048
|
39 |
-
|
40 |
-
# Build the model architecture.
|
41 |
model = MultiTaskModel(resnet, feature_dim, num_obj_classes)
|
42 |
model.to(device)
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
filename = "Yolloplusclassproject_weights.pth" # The state dict file you uploaded
|
47 |
weights_path = hf_hub_download(repo_id=repo_id, filename=filename)
|
48 |
-
|
49 |
-
# Load the state dict and update the model.
|
50 |
state_dict = torch.load(weights_path, map_location="cpu")
|
51 |
model.load_state_dict(state_dict)
|
52 |
model.eval()
|
53 |
|
54 |
########################################
|
55 |
-
# 3.
|
56 |
########################################
|
57 |
-
#
|
58 |
-
|
59 |
-
|
60 |
-
#
|
61 |
-
idx_to_obj_label = {
|
|
|
62 |
bin_label_names = ["AI-Generated", "Real"]
|
63 |
|
64 |
-
# Define the validation transforms (must match those used during training)
|
65 |
val_transforms = transforms.Compose([
|
66 |
transforms.Resize(256),
|
67 |
transforms.CenterCrop(224),
|
@@ -74,13 +65,8 @@ val_transforms = transforms.Compose([
|
|
74 |
# 4. Define the Inference Function
|
75 |
########################################
|
76 |
def predict_image(img: Image.Image) -> str:
|
77 |
-
"""
|
78 |
-
Takes an uploaded PIL image, processes it, and returns the model's prediction.
|
79 |
-
"""
|
80 |
-
# Ensure the image is in RGB mode.
|
81 |
img = img.convert("RGB")
|
82 |
-
|
83 |
-
img_tensor = val_transforms(img).unsqueeze(0).to(device) # Shape: [1, 3, 224, 224]
|
84 |
with torch.no_grad():
|
85 |
obj_logits, bin_logits = model(img_tensor)
|
86 |
obj_pred = torch.argmax(obj_logits, dim=1).item()
|
@@ -97,11 +83,7 @@ demo = gr.Interface(
|
|
97 |
inputs=gr.Image(type="pil"),
|
98 |
outputs="text",
|
99 |
title="Multi-Task Image Classifier",
|
100 |
-
description=
|
101 |
-
"Upload an image to receive two predictions:\n"
|
102 |
-
"1) The primary object in the image,\n"
|
103 |
-
"2) Whether the image is AI-generated or Real."
|
104 |
-
)
|
105 |
)
|
106 |
|
107 |
if __name__ == "__main__":
|
|
|
5 |
import torchvision.transforms as transforms
|
6 |
from PIL import Image
|
7 |
from huggingface_hub import hf_hub_download
|
8 |
+
import json
|
9 |
|
10 |
########################################
|
11 |
# 1. Define the Model Architecture
|
|
|
14 |
def __init__(self, backbone, feature_dim, num_obj_classes):
|
15 |
super(MultiTaskModel, self).__init__()
|
16 |
self.backbone = backbone
|
|
|
17 |
self.obj_head = nn.Linear(feature_dim, num_obj_classes)
|
|
|
18 |
self.bin_head = nn.Linear(feature_dim, 2)
|
19 |
|
20 |
def forward(self, x):
|
|
|
26 |
########################################
|
27 |
# 2. Reconstruct the Model and Load Weights
|
28 |
########################################
|
29 |
+
num_obj_classes = 494 # Make sure this matches your training
|
|
|
|
|
30 |
device = torch.device("cpu")
|
31 |
|
|
|
32 |
resnet = models.resnet50(pretrained=False)
|
33 |
+
resnet.fc = nn.Identity()
|
34 |
feature_dim = 2048
|
|
|
|
|
35 |
model = MultiTaskModel(resnet, feature_dim, num_obj_classes)
|
36 |
model.to(device)
|
37 |
|
38 |
+
repo_id = "Abdu07/multitask-model"
|
39 |
+
filename = "Yolloplusclassproject_weights.pth"
|
|
|
40 |
weights_path = hf_hub_download(repo_id=repo_id, filename=filename)
|
|
|
|
|
41 |
state_dict = torch.load(weights_path, map_location="cpu")
|
42 |
model.load_state_dict(state_dict)
|
43 |
model.eval()
|
44 |
|
45 |
########################################
|
46 |
+
# 3. Load Label Mapping and Define Transforms
|
47 |
########################################
|
48 |
+
# Load the saved mapping from JSON
|
49 |
+
with open("obj_label_mapping.json", "r") as f:
|
50 |
+
obj_label_to_idx = json.load(f)
|
51 |
+
# Create the inverse mapping
|
52 |
+
idx_to_obj_label = {v: k for k, v in obj_label_to_idx.items()}
|
53 |
+
|
54 |
bin_label_names = ["AI-Generated", "Real"]
|
55 |
|
|
|
56 |
val_transforms = transforms.Compose([
|
57 |
transforms.Resize(256),
|
58 |
transforms.CenterCrop(224),
|
|
|
65 |
# 4. Define the Inference Function
|
66 |
########################################
|
67 |
def predict_image(img: Image.Image) -> str:
|
|
|
|
|
|
|
|
|
68 |
img = img.convert("RGB")
|
69 |
+
img_tensor = val_transforms(img).unsqueeze(0).to(device)
|
|
|
70 |
with torch.no_grad():
|
71 |
obj_logits, bin_logits = model(img_tensor)
|
72 |
obj_pred = torch.argmax(obj_logits, dim=1).item()
|
|
|
83 |
inputs=gr.Image(type="pil"),
|
84 |
outputs="text",
|
85 |
title="Multi-Task Image Classifier",
|
86 |
+
description="Upload an image to receive two predictions:\n1) The primary object in the image,\n2) Whether the image is AI-generated or Real."
|
|
|
|
|
|
|
|
|
87 |
)
|
88 |
|
89 |
if __name__ == "__main__":
|