Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,17 +1,15 @@
|
|
1 |
-
import
|
2 |
import torch
|
|
|
3 |
from torch.utils.data import Dataset, DataLoader
|
4 |
-
from PIL import Image
|
5 |
-
import torchvision
|
6 |
from torchvision import transforms
|
|
|
7 |
import xml.etree.ElementTree as ET
|
8 |
import torch.optim as optim
|
9 |
-
|
10 |
-
import gradio as gr
|
11 |
-
|
12 |
-
# Ensure device is set to GPU if available
|
13 |
-
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
14 |
|
|
|
|
|
15 |
class FaceMaskDataset(Dataset):
|
16 |
def __init__(self, images_dir, annotations_dir, transform=None, resize=(800, 800)):
|
17 |
self.images_dir = images_dir
|
@@ -26,21 +24,17 @@ class FaceMaskDataset(Dataset):
|
|
26 |
def __getitem__(self, idx):
|
27 |
image_path = os.path.join(self.images_dir, self.image_files[idx])
|
28 |
image = Image.open(image_path).convert("RGB")
|
29 |
-
|
30 |
-
# Resize the image to a fixed size, while maintaining aspect ratio
|
31 |
image = image.resize(self.resize)
|
32 |
|
33 |
-
# Handle both .jpg and .png files
|
34 |
annotation_path = os.path.join(self.annotations_dir, self.image_files[idx].replace(".jpg", ".xml").replace(".png", ".xml"))
|
35 |
|
36 |
if not os.path.exists(annotation_path):
|
37 |
print(f"Warning: Annotation file {annotation_path} does not exist. Skipping image {self.image_files[idx]}.")
|
38 |
-
return None, None # Return
|
39 |
|
40 |
boxes, labels = self.load_annotations(annotation_path)
|
41 |
-
|
42 |
if boxes is None or labels is None:
|
43 |
-
return None, None # Skip
|
44 |
|
45 |
target = {'boxes': boxes, 'labels': labels}
|
46 |
|
@@ -63,66 +57,119 @@ class FaceMaskDataset(Dataset):
|
|
63 |
xmax = float(bndbox.find('xmax').text)
|
64 |
ymax = float(bndbox.find('ymax').text)
|
65 |
boxes.append([xmin, ymin, xmax, ymax])
|
66 |
-
labels.append(1 if label == "mask" else 0) #
|
67 |
|
68 |
if len(boxes) == 0 or len(labels) == 0:
|
69 |
-
return None, None # If no boxes or labels
|
70 |
|
71 |
boxes = torch.as_tensor(boxes, dtype=torch.float32)
|
72 |
labels = torch.tensor(labels, dtype=torch.int64)
|
73 |
|
74 |
return boxes, labels
|
75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
-
#
|
78 |
-
|
79 |
-
|
80 |
-
batch = [item for item in batch if item[0] is not None and item[1] is not None]
|
81 |
-
return tuple(zip(*batch))
|
82 |
|
83 |
-
|
84 |
-
|
85 |
-
model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained=True)
|
86 |
-
# Assuming 2 classes: mask and no_mask
|
87 |
-
num_classes = 2
|
88 |
-
in_features = model.roi_heads.box_predictor.cls_score.in_features
|
89 |
-
model.roi_heads.box_predictor = torchvision.models.detection.faster_rcnn.FastRCNNPredictor(in_features, num_classes)
|
90 |
-
model.to(device)
|
91 |
-
return model
|
92 |
|
93 |
-
#
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
transform = transforms.Compose([
|
100 |
-
transforms.Resize((224, 224)), # Resize all images to 224x224
|
101 |
-
transforms.ToTensor(),
|
102 |
-
])
|
103 |
|
104 |
-
|
105 |
-
image = transform(image).unsqueeze(0).to(device) # Add batch dimension
|
106 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
107 |
with torch.no_grad():
|
108 |
-
|
|
|
|
|
109 |
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
|
114 |
-
|
115 |
|
116 |
-
#
|
117 |
-
|
118 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
119 |
|
120 |
-
#
|
121 |
-
|
122 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
|
124 |
-
#
|
125 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
# Launch Gradio interface
|
128 |
iface.launch()
|
|
|
1 |
+
import gradio as gr
|
2 |
import torch
|
3 |
+
import os
|
4 |
from torch.utils.data import Dataset, DataLoader
|
|
|
|
|
5 |
from torchvision import transforms
|
6 |
+
from PIL import Image
|
7 |
import xml.etree.ElementTree as ET
|
8 |
import torch.optim as optim
|
9 |
+
from torch import nn
|
|
|
|
|
|
|
|
|
10 |
|
11 |
+
# Your model training and evaluation functions (already defined in your previous code)
|
12 |
+
# Define the custom dataset
|
13 |
class FaceMaskDataset(Dataset):
|
14 |
def __init__(self, images_dir, annotations_dir, transform=None, resize=(800, 800)):
|
15 |
self.images_dir = images_dir
|
|
|
24 |
def __getitem__(self, idx):
|
25 |
image_path = os.path.join(self.images_dir, self.image_files[idx])
|
26 |
image = Image.open(image_path).convert("RGB")
|
|
|
|
|
27 |
image = image.resize(self.resize)
|
28 |
|
|
|
29 |
annotation_path = os.path.join(self.annotations_dir, self.image_files[idx].replace(".jpg", ".xml").replace(".png", ".xml"))
|
30 |
|
31 |
if not os.path.exists(annotation_path):
|
32 |
print(f"Warning: Annotation file {annotation_path} does not exist. Skipping image {self.image_files[idx]}.")
|
33 |
+
return None, None # Return None if annotation is missing
|
34 |
|
35 |
boxes, labels = self.load_annotations(annotation_path)
|
|
|
36 |
if boxes is None or labels is None:
|
37 |
+
return None, None # Skip if annotations are invalid
|
38 |
|
39 |
target = {'boxes': boxes, 'labels': labels}
|
40 |
|
|
|
57 |
xmax = float(bndbox.find('xmax').text)
|
58 |
ymax = float(bndbox.find('ymax').text)
|
59 |
boxes.append([xmin, ymin, xmax, ymax])
|
60 |
+
labels.append(1 if label == "mask" else 0) # "mask" = 1, "no_mask" = 0
|
61 |
|
62 |
if len(boxes) == 0 or len(labels) == 0:
|
63 |
+
return None, None # If no boxes or labels, return None
|
64 |
|
65 |
boxes = torch.as_tensor(boxes, dtype=torch.float32)
|
66 |
labels = torch.tensor(labels, dtype=torch.int64)
|
67 |
|
68 |
return boxes, labels
|
69 |
|
70 |
+
# Model Training Loop (referred to from previous code)
|
71 |
+
def train_model(model, train_loader, val_loader, optimizer, num_epochs=10):
|
72 |
+
for epoch in range(num_epochs):
|
73 |
+
# Training loop
|
74 |
+
running_loss = 0.0
|
75 |
+
model.train()
|
76 |
+
for images, targets in train_loader:
|
77 |
+
if images is None or targets is None:
|
78 |
+
continue # Skip invalid images/annotations
|
79 |
|
80 |
+
# Move data to device
|
81 |
+
images = [image.to(device) for image in images]
|
82 |
+
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
|
|
|
|
|
83 |
|
84 |
+
optimizer.zero_grad()
|
85 |
+
loss_dict = model(images, targets)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
86 |
|
87 |
+
# Calculate total loss
|
88 |
+
total_loss = sum(loss for loss in loss_dict.values())
|
89 |
+
total_loss.backward()
|
90 |
+
optimizer.step()
|
91 |
+
|
92 |
+
running_loss += total_loss.item()
|
|
|
|
|
|
|
|
|
93 |
|
94 |
+
print(f"Epoch {epoch+1}/{num_epochs}, Loss: {running_loss / len(train_loader)}")
|
|
|
95 |
|
96 |
+
# Evaluate after every epoch
|
97 |
+
val_loss = evaluate_model(model, val_loader)
|
98 |
+
print(f"Validation Loss: {val_loss}")
|
99 |
+
|
100 |
+
# Validation function
|
101 |
+
def evaluate_model(model, val_loader):
|
102 |
+
model.eval()
|
103 |
+
running_loss = 0.0
|
104 |
with torch.no_grad():
|
105 |
+
for images, targets in val_loader:
|
106 |
+
if images is None or targets is None:
|
107 |
+
continue # Skip invalid data
|
108 |
|
109 |
+
# Move data to device
|
110 |
+
images = [image.to(device) for image in images]
|
111 |
+
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
|
112 |
|
113 |
+
loss_dict = model(images, targets)
|
114 |
|
115 |
+
# Calculate total loss
|
116 |
+
total_loss = sum(loss for loss in loss_dict.values())
|
117 |
+
running_loss += total_loss.item()
|
118 |
+
|
119 |
+
return running_loss / len(val_loader)
|
120 |
+
|
121 |
+
# Function to upload dataset and start training
|
122 |
+
def train_on_uploaded_data(train_data, val_data):
|
123 |
+
# Save the uploaded dataset (files)
|
124 |
+
train_data_path = "train_data.zip"
|
125 |
+
val_data_path = "val_data.zip"
|
126 |
|
127 |
+
# Unzip and prepare directories (assuming you upload zip files for simplicity)
|
128 |
+
with open(train_data.name, 'wb') as f:
|
129 |
+
f.write(train_data.read())
|
130 |
+
with open(val_data.name, 'wb') as f:
|
131 |
+
f.write(val_data.read())
|
132 |
+
|
133 |
+
# Extract zip files
|
134 |
+
os.system(f"unzip {train_data_path} -d ./train/")
|
135 |
+
os.system(f"unzip {val_data_path} -d ./val/")
|
136 |
+
|
137 |
+
# Load datasets
|
138 |
+
train_dataset = FaceMaskDataset(
|
139 |
+
images_dir="train/images",
|
140 |
+
annotations_dir="train/annotations",
|
141 |
+
transform=transforms.Compose([transforms.Resize((224, 224)), transforms.ToTensor()])
|
142 |
+
)
|
143 |
+
val_dataset = FaceMaskDataset(
|
144 |
+
images_dir="val/images",
|
145 |
+
annotations_dir="val/annotations",
|
146 |
+
transform=transforms.Compose([transforms.Resize((224, 224)), transforms.ToTensor()])
|
147 |
+
)
|
148 |
+
|
149 |
+
# Dataloaders
|
150 |
+
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True, collate_fn=collate_fn)
|
151 |
+
val_loader = DataLoader(val_dataset, batch_size=32, shuffle=False, collate_fn=collate_fn)
|
152 |
|
153 |
+
# Train the model
|
154 |
+
model = get_model(num_classes=2) # Assuming you have a model function
|
155 |
+
model.to(device)
|
156 |
+
optimizer = optim.SGD(model.parameters(), lr=0.005, momentum=0.9, weight_decay=0.0005)
|
157 |
+
|
158 |
+
# Train the model and return feedback
|
159 |
+
train_model(model, train_loader, val_loader, optimizer, num_epochs=10)
|
160 |
+
|
161 |
+
return "Training completed and model saved."
|
162 |
+
|
163 |
+
# Create Gradio Interface
|
164 |
+
iface = gr.Interface(
|
165 |
+
fn=train_on_uploaded_data,
|
166 |
+
inputs=[
|
167 |
+
gr.File(label="Upload Train Dataset (ZIP)"),
|
168 |
+
gr.File(label="Upload Validation Dataset (ZIP)")
|
169 |
+
],
|
170 |
+
outputs=gr.Textbox(label="Training Status"),
|
171 |
+
live=True
|
172 |
+
)
|
173 |
|
174 |
# Launch Gradio interface
|
175 |
iface.launch()
|