Comparison

Deep ResNet Scaling with Bottleneck Blocks

Bottleneck blocks support deeper ResNet configurations while controlling computational complexity:

  • ResNet-50: Replaces each two-layer block of the 34-layer baseline with a three-layer bottleneck block, uses projection shortcuts (Option B) when dimensions increase, and requires 3.8×1093.8 \times 10^9 FLOPs.
  • ResNet-101 and ResNet-152: Stack additional three-layer bottleneck blocks, especially at feature-map sizes 28×2828 \times 28 and 14×1414 \times 14. ResNet-152 requires 11.3×10911.3 \times 10^9 FLOPs, less than VGG-16 at 15.3×10915.3 \times 10^9 FLOPs and VGG-19 at 19.6×10919.6 \times 10^9 FLOPs.

Across these bottleneck ResNet configurations, increasing depth avoids the degradation problem and improves accuracy.

0

1

Updated 2026-09-12

Tags

Prep Sessions

Foundational Deep Learning Architectures: Transformers and Residual Networks @ University of Michigan - Ann Arbor

Ch.3 Deep Residual Network Architecture - Foundational Deep Learning Architectures: Transformers and Residual Networks @ University of Michigan - Ann Arbor

Deeper Bottleneck Architectures - Foundational Deep Learning Architectures: Transformers and Residual Networks @ University of Michigan - Ann Arbor

Ch.4 Residual Network Experiments and Applications - Foundational Deep Learning Architectures: Transformers and Residual Networks @ University of Michigan - Ann Arbor

ImageNet Classification and Model Variations - Foundational Deep Learning Architectures: Transformers and Residual Networks @ University of Michigan - Ann Arbor

Related