Common CNN Architecture

(and their benchmarks and usecase)

Name Parameters (M) Speed Accuracy (ImageNet Top-1)
AlexNet 60 Slow 56.5%
VGG-16 138 Very Slow ~71% (est.)
GoogLeNet (Inception V1) 5 Fast ~69% (est.)
ResNet-50 26 Moderate ~80.3%
ResNet-101 44 Moderate-Slow ~80.3%
Inception-ResNet-V2 55 Slow 80.3%
MobileNetV3-Large 5.5 Very Fast 75.27%
EfficientNetV2-S 21.5 Moderate-Fast 84.23%
ConvNeXt-Tiny 28.6 Moderate 82.52%

Name Use Case
AlexNet Historical, early object detection
VGG-16 General high-accuracy detection
GoogLeNet (Inception V1) Lightweight detection
ResNet-50 High-accuracy general detection
ResNet-101 High-accuracy complex datasets
Inception-ResNet-V2 High-accuracy complex tasks
MobileNetV3-Large Real-time detection on mobile/edge devices
EfficientNetV2-S Versatile detection across domains
ConvNeXt-Tiny Specialized high-accuracy tasks (e.g., infrared, remote sensing)