Spaces:
Sleeping
Sleeping
File size: 6,488 Bytes
22ae476 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 |
import gradio as gr
from huggingface_hub import hf_hub_download
import torch
import torch.nn as nn
from torchvision import transforms
class SimpleResidualBlock(nn.Module):
def __init__(self, in_channels, out_channels, set_stride=False):
super().__init__()
stride = 2 if in_channels != out_channels and set_stride else 1
self.conv1 = nn.LazyConv2d(
out_channels,
kernel_size=3,
padding="same" if stride == 1 else 1,
stride=stride,
)
self.conv2 = nn.LazyConv2d(out_channels, kernel_size=3, padding="same")
self.bn1 = nn.LazyBatchNorm2d()
self.bn2 = nn.LazyBatchNorm2d()
self.relu = nn.ReLU()
if in_channels != out_channels:
self.residual = nn.Sequential(
nn.LazyConv2d(out_channels, kernel_size=1, stride=stride),
nn.LazyBatchNorm2d(),
)
else:
self.residual = nn.Identity()
def forward(self, x):
out = self.relu(self.bn1(self.conv1(x)))
out = self.bn2(self.conv2(out))
out += self.residual(x)
out = self.relu(out)
return out
class BottleneckResidualBlock(nn.Module):
def __init__(
self, in_channels, out_channels, identity_mapping=False, set_stride=False
):
super().__init__()
stride = 2 if in_channels != out_channels and set_stride else 1
self.conv1 = nn.LazyConv2d(
out_channels,
kernel_size=1,
padding="same" if stride == 1 else 0,
stride=stride,
)
self.conv2 = nn.LazyConv2d(out_channels, kernel_size=3, padding="same")
self.conv3 = nn.LazyConv2d(out_channels * 4, kernel_size=1, padding="same")
self.bn1 = nn.LazyBatchNorm2d()
self.bn2 = nn.LazyBatchNorm2d()
self.bn3 = nn.LazyBatchNorm2d()
self.relu = nn.ReLU()
if in_channels != out_channels or not identity_mapping:
self.residual = nn.Sequential(
nn.LazyConv2d(out_channels * 4, kernel_size=1, stride=stride),
nn.LazyBatchNorm2d(),
)
else:
self.residual = nn.Identity()
def forward(self, x):
out = self.relu(self.bn1(self.conv1(x)))
out = self.relu(self.bn2(self.conv2(out)))
out = self.bn3(self.conv3(out))
out += self.residual(x)
out = self.relu(out)
return out
RESNET_18 = [2, 2, 2, 2]
RESNET_34 = [3, 4, 6, 3]
RESNET_50 = [3, 4, 6, 3]
RESNET_101 = [3, 4, 23, 3]
RESNET_152 = [3, 8, 36, 3]
class ResNet(nn.Module):
def __init__(self, arch=RESNET_18, block="simple", num_classes=256):
super().__init__()
self.conv1 = nn.Sequential(
nn.LazyConv2d(64, kernel_size=7, stride=2, padding=3),
nn.LazyBatchNorm2d(),
nn.ReLU(),
)
self.maxpool = nn.MaxPool2d(3, stride=2, padding=1)
self.conv2 = self._make_layer(64, 64, arch[0], set_stride=False, block=block)
self.conv3 = self._make_layer(64, 128, arch[1], block=block)
self.conv4 = self._make_layer(128, 256, arch[2], block=block)
self.conv5 = self._make_layer(256, 512, arch[3], block=block)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.flatten = nn.Flatten()
self.fc = nn.LazyLinear(num_classes)
def _make_layer(
self, in_channels, out_channels, num_blocks, set_stride=True, block="simple"
):
"""Block is either 'simple' or 'bottleneck'"""
layers = []
for i in range(num_blocks):
layers.append(
SimpleResidualBlock(in_channels, out_channels, set_stride=set_stride)
if block == "simple"
else BottleneckResidualBlock(
in_channels if i == 0 else out_channels * 4,
out_channels,
set_stride=set_stride,
)
)
set_stride = False
return nn.Sequential(*layers)
def forward(self, x):
out = self.conv1(x)
out = self.maxpool(self.conv2(out))
out = self.conv3(out)
out = self.conv4(out)
out = self.conv5(out)
out = self.avgpool(out)
out = self.flatten(out)
out = self.fc(out)
return out
def _init_weights(module):
# Initlize weights with glorot uniform
if isinstance(module, nn.Conv2d):
nn.init.xavier_uniform_(module.weight)
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Linear):
nn.init.xavier_uniform_(module.weight)
nn.init.zeros_(module.bias)
class ImageClassifier:
def __init__(self, checkpoint_path):
self.checkpoint_path = checkpoint_path
self.model = self.load_model(checkpoint_path)
self.transform = self.get_transform((244, 244))
self.labels = [
"airplane",
"automobile",
"bird",
"cat",
"deer",
"dog",
"frog",
"horse",
"ship",
"truck",
]
def load_model(self, checkpoint_path):
classifier = ResNet(
arch=RESNET_18,
block="simple",
num_classes=10,
)
classifier.load_state_dict(torch.load(checkpoint_path))
classifier = classifier.cpu()
classifier.eval()
return classifier
def get_transform(self, img_shape):
preprocess_transform = transforms.Compose(
[
transforms.Resize(img_shape),
transforms.ToTensor(),
]
)
return preprocess_transform
def predict(self, image):
image_tensor = self.transform(image).unsqueeze(0)
with torch.no_grad():
logits = self.model(image_tensor)
probs = logits.softmax(dim=1)[0]
return {label: prob.item() for label, prob in zip(self.labels, probs)}
def classify(self, input_image):
return self.predict(input_image)
def classify(input_image):
return classifier.classify(input_image)
checkpoint_path = hf_hub_download(
repo_id="SatwikKambham/resnet18-cifar10",
filename="model.pt",
)
classifier = ImageClassifier(checkpoint_path)
iface = gr.Interface(
classify,
inputs=[
gr.Image(label="Input Image", type="pil"),
],
outputs=gr.Label(num_top_classes=3),
)
iface.launch()
|