AI Paper Atlas

Research map / Detection and segmentation / CNN features

Beyond MACs: Hardware Efficient Architecture Design for Vision Backbones

Moritz Nottebaum, Matteo Dunnhofer, Christian Micheloni

IJCV 2026 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 1 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
add fused MBConv in place of unfused MBConv (ablation)LowFormer-B1 · ImageNet-1Ktop-1 accuracy: 79.1 -> 79.9Improved
add fused MBConv in place of unfused MBConv (ablation)LowFormer-B1 · ImageNet-1KGPU throughput: 3558 -> 4237Improved
add Lowtention in place of ReLU linear attention (ablation)LowFormer-B1 · ImageNet-1Ktop-1 accuracy: 79.6 -> 79.9Improved
add downscaling of feature maps in Lowtention (ablation)LowFormer-B1 · ImageNet-1Ktop-1 accuracy: 79.9 -> 79.9No clear effect
add downscaling of feature maps in Lowtention (ablation)LowFormer-B1 · ImageNet-1KTX2 latency: 13.4 -> 11.7Improved
add channel compression in Lowtention (ablation)LowFormer-B1 · ImageNet-1KGPU throughput: 3921 -> 4237Improved
add Lowtention in place of original MHSA (ablation)LowFormer-B1 · ImageNet-1Ktop-1 accuracy: 79.8 -> 79.9Improved
add Lowtention in place of original MHSA at 1024x1024 input (ablation)LowFormer-B1 · ImageNet-1K at 1024x1024TX2 latency: 220% higher latency without downsamplingImproved
remove MLP from LowFormer-B1 (ablation)LowFormer-B1 (B1_mlpless) · ImageNet-1Ktop-1 accuracy: 79.9 -> 78.8Hurt
remove MLP and reduce depth (shallow) (ablation)LowFormer-B1 · ImageNet-1Ktop-1 accuracy: 78.8 -> 77.2Hurt
remove Lowtention (conv only) from B1.5 shallow (ablation)LowFormer-B1.5 · ImageNet-1Ktop-1 accuracy: 79.7 -> 78.8Hurt
remove MLP from LowFormer-B3 (edge variant E3)LowFormer-B3 · ImageNet-1KTX2 latency: 32.5 -> 25.0Improved

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