AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Model Rubik’s Cube: Twisting Resolution, Depth and Width for TinyNets

Kai Han, Yunhe Wang, Qiulin Zhang, Wei Zhang, Chunjing XU, Tong Zhang

NeurIPS 2020 · Architectures and efficiency › RNNs and normalization · approach: Efficient network compression and training · 47 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
shrink with tiny formula (r,d first, width from FLOPs) instead of inverse EfficientNet compound ruleEfficientNet-B0 at 200M FLOPs · ImageNet-100top-1 accuracy: 75.8 -> 77.6Improved
tiny formula instead of inverse compound ruleEfficientNet-B0 at 100M FLOPs · ImageNet-100top-1 accuracy: 72.1 -> 74.1Improved
tiny formula vs shrinking only resolution (ablation)EfficientNet-B0 200M · ImageNet-100top-1 accuracy: 74.9 -> 77.6Improved
tiny formula vs shrinking only width (ablation)EfficientNet-B0 200M · ImageNet-100top-1 accuracy: 77.2 -> 77.6Improved
tiny formula instead of inverse compound ruleEfficientNet-B0 at 24M FLOPs · ImageNet-1000top-1 accuracy: 56.7 -> 59.9Improved
tiny formula instead of inverse compound ruleEfficientNet-B0 at 339M vs 387M · ImageNet-1000top-1 accuracy: 76.7 -> 76.8No clear effect
add RandAugmentTinyNet-A · ImageNet-1000top-1 accuracy: 76.8 -> 77.7Improved
add RandAugmentEfficientNet-B0 · ImageNet-1000top-1 accuracy: 76.7 -> 77.7Improved
tiny formula instead of EfficientNet rule on ResNet-50ResNet-50 at 1.8B FLOPs · ImageNet-100top-1 accuracy: 76.9 -> 78.2Improved
tiny formula instead of EfficientNet rule on ResNet-50ResNet-50 at 3.6B FLOPs · ImageNet-100top-1 accuracy: 78.3 -> 79.3Improved
swap EfficientNet-B-3 backbone for TinyNet-DSSDLite · MS COCOmAP: 17.1 -> 19.2Improved
tiny formula vs width-only shrinkGhostNet at 300M FLOPs · ImageNet-1000top-1 accuracy: 76.4 -> 77.0Improved

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