d22cs051 commited on
Commit
f217fdb
1 Parent(s): 25ff115

uploading app and initial commit

Browse files
.gitattributes CHANGED
@@ -32,3 +32,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
32
  *.zip filter=lfs diff=lfs merge=lfs -text
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
32
  *.zip filter=lfs diff=lfs merge=lfs -text
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
35
+ examples/Image_4.jpg filter=lfs diff=lfs merge=lfs -text
36
+ examples/Image_89.jpg filter=lfs diff=lfs merge=lfs -text
app.py ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ### 1. Imports and class names setup ###
2
+ import gradio as gr
3
+ import os
4
+ import torch
5
+ from torchvision import transforms
6
+
7
+ from models import get_mobilenet_v2_model, get_resnet_18_model, get_vgg_16_model
8
+ from timeit import default_timer as timer
9
+ from typing import Tuple, Dict
10
+
11
+ # Set device
12
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
13
+
14
+ # Setup class names
15
+ class_names = ["car","dragon","hourse","pegasus","ship","t-rex","tree"]
16
+
17
+ ### 2. Model and transforms preparation ###
18
+
19
+ # Create EffNetB2 model
20
+ img_transforms = transforms.Compose(
21
+ [
22
+ transforms.Resize(size=(224, 224)),
23
+ transforms.ToTensor(),
24
+ ]
25
+ )
26
+
27
+ model_name_to_fn = {
28
+ "mobilenet_v2": get_mobilenet_v2_model,
29
+ "resnet_18": get_resnet_18_model,
30
+ "vgg_16": get_vgg_16_model,
31
+ }
32
+ model_name_to_path = {
33
+ "mobilenet_v2": "mobilenet_v2.pth",
34
+ "resnet_18": "resnet_18.pth",
35
+ "vgg_16": "vgg_16.pt",
36
+ }
37
+
38
+ ### 3. Predict function ###
39
+
40
+
41
+ # Create predict function
42
+ def predict(img, model_name: str,) -> Tuple[Dict, float]:
43
+ """
44
+ Desc: Transforms and performs a prediction on img and returns prediction and time taken.
45
+ Args:
46
+ model_name (str): Name of the model to use for prediction.
47
+ img (PIL.Image): Image to perform prediction on.
48
+ Returns:
49
+ Tuple[Dict, float]: Tuple containing a dictionary of prediction labels and probabilities and the time taken to perform the prediction.
50
+ """
51
+ # Start the timer
52
+ start_time = timer()
53
+
54
+ # Get the model function based on the model name
55
+ model_fn = model_name_to_fn[model_name]
56
+ model_path = model_name_to_path[model_name]
57
+
58
+ # Create the model and load its weights
59
+ model = model_fn().to(device)
60
+ model.load_state_dict(
61
+ torch.load(f"./models/{model_name}.pth", map_location=torch.device(device=device))
62
+ )
63
+
64
+ # Put model into evaluation mode and turn on inference mode
65
+ model.eval()
66
+ with torch.inference_mode():
67
+ # Transform the target image and add a batch dimension
68
+ img = img_transforms(img).unsqueeze(0).to(device)
69
+
70
+ # Pass the transformed image through the model and turn the prediction logits into prediction probabilities
71
+ pred_probs = torch.softmax(model(img), dim=1)
72
+
73
+ # Create a prediction label and prediction probability dictionary for each prediction class (this is the required format for Gradio's output parameter)
74
+ pred_labels_and_probs = {
75
+ class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))
76
+ }
77
+
78
+ # Calculate the prediction time
79
+ pred_time = round(timer() - start_time, 5)
80
+
81
+ # Return the prediction dictionary and prediction time
82
+ return pred_labels_and_probs, pred_time
83
+
84
+
85
+ ### 4. Gradio app ###
86
+
87
+ # Create title, description and article strings
88
+ title = "SketchRec Mini ✍🏻"
89
+ description = "An Mutimodel Sketch Recognition App 🎨"
90
+ article = ""
91
+
92
+ # Create examples list from "examples/" directory
93
+ example_list = [["examples/" + example] for example in os.listdir("examples")]
94
+
95
+ # Create the Gradio demo
96
+ model_selection_dropdown = gr.components.Dropdown(
97
+ choices=list(model_name_to_fn.keys()), label="Select a model",
98
+ value="mobilenet_v2"
99
+ )
100
+
101
+ demo = gr.Interface(
102
+ fn=predict, # mapping function from input to output
103
+ inputs=[gr.Image(type="pil"),model_selection_dropdown], # what are the inputs?
104
+ outputs=[
105
+ gr.Label(num_top_classes=7, label="Predictions"), # what are the outputs?
106
+ gr.Number(label="Prediction time (s)"),
107
+ ], # our fn has two outputs, therefore we have two outputs
108
+ # Create examples list from "examples/" directory
109
+ examples=example_list,
110
+ title=title,
111
+ description=description,
112
+ article=article,
113
+ )
114
+
115
+ # Launch the demo!
116
+ demo.launch(
117
+ debug=True,
118
+ )
examples/Image_105.jpg ADDED
examples/Image_18.jpg ADDED
examples/Image_28.jpg ADDED
examples/Image_4.jpg ADDED

Git LFS Details

  • SHA256: 69490cc9f80a63261d8aac25217fa7683bac5a8e8fe07e972b7d7fb2467b80c9
  • Pointer size: 132 Bytes
  • Size of remote file: 1.85 MB
examples/Image_58.jpg ADDED
examples/Image_67.jpg ADDED
examples/Image_89.jpg ADDED

Git LFS Details

  • SHA256: a24199bbdca306ccd599b2cf766bcd9c2224e1a540f959e857379292fb27cd1a
  • Pointer size: 132 Bytes
  • Size of remote file: 1.07 MB
models.py ADDED
@@ -0,0 +1,183 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from typing import List
2
+ import torch
3
+ from torch import nn
4
+ import numpy as np
5
+ from torchvision import models
6
+ from torchvision.models import ResNet18_Weights,ResNet50_Weights,VGG16_Weights,MobileNet_V2_Weights
7
+
8
+
9
+ class EarlyStopping:
10
+ def __init__(self, tolerance=5, min_delta=0):
11
+
12
+ self.tolerance = tolerance
13
+ self.min_delta = min_delta
14
+ self.counter = 0
15
+ self.early_stop = False
16
+
17
+ def __call__(self, train_loss, validation_loss):
18
+ if (validation_loss - train_loss) > self.min_delta:
19
+ self.counter +=1
20
+ if self.counter >= self.tolerance:
21
+ self.early_stop = True
22
+
23
+
24
+
25
+ class Resnet18(nn.Module):
26
+ def __init__(self,out_shape:int = 1000) -> None:
27
+ super().__init__()
28
+ self.resnet = models.resnet18()
29
+ self.resnet.fc = nn.Linear(512,out_shape)
30
+
31
+ def forward(self,x):
32
+ return self.resnet(x)
33
+
34
+
35
+ class PretrainedResnet18(nn.Module):
36
+ def __init__(self,out_shape:int = 1000) -> None:
37
+ super().__init__()
38
+ self.resnet = models.resnet18(weights=ResNet18_Weights.DEFAULT)
39
+
40
+ # freeze all layers except last fc layer
41
+ for parms in self.resnet.parameters():
42
+ parms.requires_grad = False
43
+
44
+ self.resnet.fc = nn.Linear(512,out_shape)
45
+
46
+ def forward(self,x):
47
+ return self.resnet(x)
48
+
49
+
50
+ class Resnet50(nn.Module):
51
+ def __init__(self,out_shape:int = 1000) -> None:
52
+ super().__init__()
53
+ self.resnet = models.resnet50()
54
+ self.resnet.fc = nn.Linear(2048,out_shape)
55
+
56
+ def forward(self,x):
57
+ return self.resnet(x)
58
+
59
+
60
+ class PretrainedResnet50(nn.Module):
61
+ def __init__(self,out_shape:int = 1000) -> None:
62
+ super().__init__()
63
+ self.resnet = models.resnet50(weights=ResNet50_Weights.DEFAULT)
64
+
65
+ # freeze all layers except last fc layer
66
+ for parms in self.resnet.parameters():
67
+ parms.requires_grad = False
68
+
69
+ self.resnet.fc = nn.Linear(2048,out_shape)
70
+
71
+ def forward(self,x):
72
+ return self.resnet(x)
73
+
74
+
75
+ class EfficentNetB0(nn.Module):
76
+ def __init__(self,out_shape:int = 1000) -> None:
77
+ super().__init__()
78
+ self.effnet = models.efficientnet_b0()
79
+ self.effnet.classifier = nn.Linear(1280,out_shape)
80
+
81
+ def forward(self,x):
82
+ return self.effnet(x)
83
+
84
+
85
+ class MobileNetV2(nn.Module):
86
+ def __init__(self,out_shape:int = 1000) -> None:
87
+ super().__init__()
88
+ self.mobilenet = models.mobilenet_v2()
89
+ self.mobilenet.classifier[1] = nn.Linear(1280,out_shape)
90
+
91
+ def forward(self,x):
92
+ return self.mobilenet(x)
93
+
94
+ class PretrainedMobileNetV2(nn.Module):
95
+ def __init__(self,out_shape:int = 1000) -> None:
96
+ super().__init__()
97
+ self.mobilenet = models.mobilenet_v2(weights=MobileNet_V2_Weights.DEFAULT)
98
+
99
+ # freeze all layers except last fc layer
100
+ for parms in self.mobilenet.parameters():
101
+ parms.requires_grad = False
102
+
103
+ self.mobilenet.classifier = nn.Sequential(
104
+ nn.Dropout(p=0.2, inplace=False),
105
+ nn.Linear(in_features=1280, out_features=1000, bias=True)
106
+ )
107
+
108
+ def forward(self,x):
109
+ return self.mobilenet(x)
110
+
111
+
112
+ class VGG16(nn.Module):
113
+ def __init__(self,out_shape:int = 1000) -> None:
114
+ super().__init__()
115
+ self.vgg = models.vgg16()
116
+ self.vgg.classifier = nn.Sequential(
117
+ nn.Linear(in_features=25088, out_features=4096, bias=True),
118
+ nn.ReLU(inplace=True),
119
+ nn.Dropout(p=0.5, inplace=False),
120
+ nn.Linear(in_features=4096, out_features=4096, bias=True),
121
+ nn.ReLU(inplace=True),
122
+ nn.Dropout(p=0.5, inplace=False),
123
+ nn.Linear(in_features=4096, out_features=out_shape, bias=True),
124
+ )
125
+
126
+ def forward(self,x):
127
+ return self.vgg(x)
128
+
129
+
130
+ class PretrainedVGG16(nn.Module):
131
+ def __init__(self,out_shape:int = 1000) -> None:
132
+ super().__init__()
133
+ self.vgg = models.vgg16(weights=VGG16_Weights.DEFAULT)
134
+
135
+ # freeze all layers except last clf layer
136
+ for parms in self.vgg.parameters():
137
+ parms.requires_grad = False
138
+
139
+ self.vgg.classifier = nn.Sequential(
140
+ nn.Linear(in_features=25088, out_features=4096, bias=True),
141
+ nn.ReLU(inplace=True),
142
+ nn.Dropout(p=0.5, inplace=False),
143
+ nn.Linear(in_features=4096, out_features=4096, bias=True),
144
+ nn.ReLU(inplace=True),
145
+ nn.Dropout(p=0.5, inplace=False),
146
+ nn.Linear(in_features=4096, out_features=out_shape, bias=True),
147
+ )
148
+
149
+ def forward(self,x):
150
+ return self.vgg(x)
151
+
152
+
153
+
154
+ class VIT(nn.Module):
155
+ def __init__(self,out_shape:int = 1000) -> None:
156
+ super().__init__()
157
+ self.vit = models.vit_b_16()
158
+ self.vit.head = nn.Linear(768,out_shape)
159
+
160
+ def forward(self,x):
161
+ return self.vit(x)
162
+
163
+
164
+
165
+ # functions to get models
166
+ def get_resnet_18_model():
167
+ model = Resnet18(out_shape=7)
168
+ return model
169
+
170
+
171
+ def get_resnet_50_model():
172
+ model = Resnet50(out_shape=7)
173
+ return model
174
+
175
+
176
+ def get_vgg_16_model():
177
+ model = VGG16(out_shape=7)
178
+ return model
179
+
180
+
181
+ def get_mobilenet_v2_model():
182
+ model = MobileNetV2(out_shape=7)
183
+ return model
models/mobilenet_v2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7597e7b6ba1183b33aca66ebf39eb73359bbc568b7293e2e62bc52cdad3f48e7
3
+ size 9263285
models/resnet_18.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ce03f8579e35fd33405cf5ac1b30fe62f87a8b47dc8c635fdaa500398bd915b2
3
+ size 44822093
models/vgg_16.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4c5f6a03dbaaa82674d76eaf0fd7fb831633d2929a25e15040ca3f05d4761f1b
3
+ size 537176517
requirements.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ torch
2
+ torchvision
3
+ gradio