CS5720 - Week 5
Slide 90 of 100
Interactive Architecture Comparison Demo
ResNet
2015
25.6M
Parameters
76.2%
ImageNet Acc
🔄 Skip Connections
Identity shortcuts enable very deep networks
DenseNet
2017
8.0M
Parameters
74.6%
ImageNet Acc
🕸️ Dense Connections
Every layer connects to every other layer
MobileNet
2017
4.2M
Parameters
70.6%
ImageNet Acc
📱 Mobile Efficiency
Depthwise separable convolutions for mobile
EfficientNet
2019
5.3M
Parameters
84.3%
ImageNet Acc
⚖️ Compound Scaling
Systematic scaling of depth, width, resolution
Performance vs Efficiency Trade-offs
Parameters (Millions) →
↑ Accuracy (%)
0
5
10
20
30
65
70
75
80
85
MobileNet
EfficientNet
DenseNet
ResNet
Interactive Comparison Controls
Accuracy
Parameters
Inference Speed
Efficiency Ratio
Architecture
Parameters
FLOPS
ImageNet Top-1
Mobile Latency
Efficiency Score
ResNet-50
25.6M
4.1B
76.2%
200ms
3.2
DenseNet-121
8.0M
2.9B
74.6%
180ms
6.8
MobileNet
4.2M
0.57B
70.6%
22ms
12.4
EfficientNet-B0
5.3M
0.39B
77.3%
28ms
15.2
← Previous
Next →
Prepared by Dr. Gorkem Kar
Modal Title
×
Modal content goes here...