AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Rethinking Bottleneck Structure for Efficient Mobile Network Design

Daquan Zhou, Qibin Hou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan

ECCV 2020 · Architectures and efficiency › RNNs and normalization · approach: Differentiable optimization layers · 304 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use sandglass block instead of bottleneck with depthwise conv in the bottleneck (ablation)Bottleneck variant (a) · ImageNettop-1 accuracy (%): 68.90 -> 74.02Improved
add a second 3x3 depthwise convolution in the bottleneck (ablation)Bottleneck variant (a) · ImageNettop-1 accuracy (%): 68.90 -> 70.11Improved
place depthwise convolutions in high-dimensional space with shortcuts between high-dimensional features (ablation)Variant (b) · ImageNettop-1 accuracy (%): 70.11 -> 74.02Improved
move depthwise conv to bottleneck positions of inverted residual block (ablation)Inverted residual block · ImageNettop-1 accuracy (%): 74.02 -> 69.26Hurt
use sandglass block instead of inverted residual block in MobileNetV2MobileNetV2 · ImageNettop-1 accuracy (%): 72.3 -> 74.02Improved
add sandglass block to DARTS search spaceDARTS · CIFAR-10test error (%): 3.11 -> 2.98Improved
add sandglass block to DARTS search spaceDARTS · CIFAR-10parameters (M): 3.25 -> 2.45Improved
add inverted residual block to DARTS search spaceDARTS · CIFAR-10test error (%): 3.11 -> 3.26Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy