Spaces:
Runtime error
Runtime error
File size: 10,316 Bytes
062cf9b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 | import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import models
import torchvision.models.video as video_models
class MobileNet3D(nn.Module):
"""Lightweight 3D CNN based on MobileNet principles"""
def __init__(self, num_classes=101, input_channels=3, dropout=0.2):
super(MobileNet3D, self).__init__()
# 3D Depthwise Separable Convolutions
self.conv1 = nn.Conv3d(input_channels, 32, kernel_size=(3, 7, 7), stride=(1, 2, 2), padding=(1, 3, 3))
self.bn1 = nn.BatchNorm3d(32)
# Depthwise separable blocks
self.dw_blocks = nn.ModuleList([
self._make_dw_block(32, 64, (1, 2, 2)),
self._make_dw_block(64, 128, (2, 2, 2)),
self._make_dw_block(128, 256, (2, 2, 2)),
self._make_dw_block(256, 512, (2, 2, 2)),
])
# Global average pooling and classifier
self.global_pool = nn.AdaptiveAvgPool3d(1)
self.dropout = nn.Dropout(dropout)
self.classifier = nn.Linear(512, num_classes)
# Initialize weights
self._initialize_weights()
def _make_dw_block(self, in_channels, out_channels, stride):
"""Create a depthwise separable 3D block"""
return nn.Sequential(
# Depthwise 3D conv
nn.Conv3d(in_channels, in_channels, kernel_size=3, stride=stride,
padding=1, groups=in_channels, bias=False),
nn.BatchNorm3d(in_channels),
nn.ReLU6(inplace=True),
# Pointwise 3D conv
nn.Conv3d(in_channels, out_channels, kernel_size=1, bias=False),
nn.BatchNorm3d(out_channels),
nn.ReLU6(inplace=True)
)
def _initialize_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv3d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.BatchNorm3d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.normal_(m.weight, 0, 0.01)
nn.init.constant_(m.bias, 0)
def forward(self, x):
# Input: [B, C, T, H, W]
x = F.relu(self.bn1(self.conv1(x)))
for block in self.dw_blocks:
x = block(x)
x = self.global_pool(x)
x = x.view(x.size(0), -1)
x = self.dropout(x)
x = self.classifier(x)
return x
class Efficient3DCNN(nn.Module):
"""Very lightweight 3D CNN for fast training"""
def __init__(self, num_classes=101, input_channels=3):
super(Efficient3DCNN, self).__init__()
# Efficient 3D feature extractor
self.features = nn.Sequential(
# Block 1
nn.Conv3d(input_channels, 32, kernel_size=(3, 7, 7), stride=(1, 2, 2), padding=(1, 3, 3)),
nn.BatchNorm3d(32),
nn.ReLU(inplace=True),
# Block 2
nn.Conv3d(32, 64, kernel_size=(3, 3, 3), stride=(2, 2, 2), padding=(1, 1, 1)),
nn.BatchNorm3d(64),
nn.ReLU(inplace=True),
# Block 3
nn.Conv3d(64, 128, kernel_size=(3, 3, 3), stride=(2, 2, 2), padding=(1, 1, 1)),
nn.BatchNorm3d(128),
nn.ReLU(inplace=True),
# Block 4 - Temporal reduction
nn.Conv3d(128, 256, kernel_size=(3, 3, 3), stride=(2, 2, 2), padding=(1, 1, 1)),
nn.BatchNorm3d(256),
nn.ReLU(inplace=True),
)
# Classifier
self.classifier = nn.Sequential(
nn.AdaptiveAvgPool3d(1),
nn.Flatten(),
nn.Dropout(0.3),
nn.Linear(256, 128),
nn.ReLU(inplace=True),
nn.Dropout(0.2),
nn.Linear(128, num_classes)
)
def forward(self, x):
x = self.features(x)
x = self.classifier(x)
return x
class ResNet2Plus1D_Light(nn.Module):
"""Lightweight (2+1)D ResNet - separates spatial and temporal convolutions"""
def __init__(self, num_classes=101):
super(ResNet2Plus1D_Light, self).__init__()
# Use pretrained R(2+1)D but make it lighter
try:
# Use smaller version if available
self.backbone = video_models.r2plus1d_18(pretrained=True)
except:
# Fallback to building our own
self.backbone = self._build_light_r2plus1d()
# Replace the final layer
if hasattr(self.backbone, 'fc'):
in_features = self.backbone.fc.in_features
self.backbone.fc = nn.Linear(in_features, num_classes)
else:
# Custom classifier
self.classifier = nn.Linear(512, num_classes)
def _build_light_r2plus1d(self):
"""Build a lightweight (2+1)D model"""
return nn.Sequential(
# Spatial conv + temporal conv blocks
nn.Conv3d(3, 32, kernel_size=(1, 7, 7), stride=(1, 2, 2), padding=(0, 3, 3)),
nn.BatchNorm3d(32),
nn.ReLU(inplace=True),
nn.Conv3d(32, 32, kernel_size=(3, 1, 1), stride=(1, 1, 1), padding=(1, 0, 0)),
nn.BatchNorm3d(32),
nn.ReLU(inplace=True),
nn.MaxPool3d(kernel_size=(1, 3, 3), stride=(1, 2, 2), padding=(0, 1, 1)),
# More (2+1)D blocks
self._make_2plus1d_block(32, 64, stride=(2, 2, 2)),
self._make_2plus1d_block(64, 128, stride=(2, 2, 2)),
self._make_2plus1d_block(128, 256, stride=(2, 2, 2)),
nn.AdaptiveAvgPool3d(1),
nn.Flatten()
)
def _make_2plus1d_block(self, in_channels, out_channels, stride=(1, 1, 1)):
"""Create a (2+1)D block"""
return nn.Sequential(
# Spatial convolution
nn.Conv3d(in_channels, out_channels, kernel_size=(1, 3, 3),
stride=(1, stride[1], stride[2]), padding=(0, 1, 1)),
nn.BatchNorm3d(out_channels),
nn.ReLU(inplace=True),
# Temporal convolution
nn.Conv3d(out_channels, out_channels, kernel_size=(3, 1, 1),
stride=(stride[0], 1, 1), padding=(1, 0, 0)),
nn.BatchNorm3d(out_channels),
nn.ReLU(inplace=True)
)
def forward(self, x):
if hasattr(self.backbone, 'fc'):
return self.backbone(x)
else:
features = self.backbone(x)
return self.classifier(features)
def get_optimized_model(model_type="efficient", num_classes=101, pretrained=True, freeze_backbone=False):
"""Get an optimized model for faster training"""
if model_type == "mobilenet3d":
print("🔧 Using MobileNet3D (lightest, fastest training)")
return MobileNet3D(num_classes=num_classes)
elif model_type == "efficient":
print("🔧 Using Efficient3DCNN (very light, good balance)")
return Efficient3DCNN(num_classes=num_classes)
elif model_type == "r2plus1d_light":
print("🔧 Using ResNet(2+1)D Light (moderate complexity)")
return ResNet2Plus1D_Light(num_classes=num_classes)
elif model_type == "r3d18_transfer":
print("🔧 Using R3D-18 with TRANSFER LEARNING (frozen backbone)")
# Load pre-trained R3D-18
model = video_models.r3d_18(pretrained=pretrained)
# Freeze all backbone layers if requested
if freeze_backbone:
print("🧊 Freezing backbone layers - only training classifier")
for name, param in model.named_parameters():
if 'fc' not in name: # Don't freeze final classifier
param.requires_grad = False
# Replace final classifier
model.fc = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(model.fc.in_features, 256),
nn.ReLU(inplace=True),
nn.Dropout(0.3),
nn.Linear(256, num_classes)
)
# Count trainable parameters
total_params = sum(p.numel() for p in model.parameters())
trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print(f"📊 Total params: {total_params:,} | Trainable: {trainable_params:,}")
return model
elif model_type == "r3d18_light":
print("🔧 Using R3D-18 with optimizations")
# Original R3D-18 but with some optimizations
model = video_models.r3d_18(pretrained=pretrained)
model.fc = nn.Sequential(
nn.Dropout(0.3),
nn.Linear(model.fc.in_features, 512),
nn.ReLU(inplace=True),
nn.Dropout(0.2),
nn.Linear(512, num_classes)
)
return model
else:
raise ValueError(f"Unknown model type: {model_type}")
def count_parameters(model):
"""Count the number of trainable parameters"""
total_params = sum(p.numel() for p in model.parameters())
trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
print(f"Total parameters: {total_params:,}")
print(f"Trainable parameters: {trainable_params:,}")
print(f"Model size: ~{total_params * 4 / 1024 / 1024:.2f} MB")
return total_params, trainable_params
def get_model_comparison():
"""Compare different model architectures"""
print("\n📊 Model Comparison:")
print("=" * 60)
models_info = {
"MobileNet3D": {"params": "~2-5M", "speed": "Fastest", "accuracy": "Good"},
"Efficient3DCNN": {"params": "~1-3M", "speed": "Very Fast", "accuracy": "Good"},
"R(2+1)D Light": {"params": "~5-10M", "speed": "Fast", "accuracy": "Better"},
"R3D-18 Original": {"params": "~33M", "speed": "Slow", "accuracy": "Best"}
}
for name, info in models_info.items():
print(f"{name:20} | Params: {info['params']:8} | Speed: {info['speed']:9} | Accuracy: {info['accuracy']}")
print("=" * 60)
print("💡 Recommendation: Start with 'efficient' or 'mobilenet3d' for faster training!") |