AI Paper Atlas

Research map / Architectures and efficiency / Quantization

DBQ: A Differentiable Branch Quantizer for Lightweight Deep Neural Networks

Hassan Dbouk, Hetul Sanghvi, Mahesh Mehendale, Naresh Shanbhag

ECCV 2020 · Architectures and efficiency › Quantization · approach: Mixed-precision quantization · 8 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 a second ternary branch to the pointwise layers (ablation)DBQ on MobileNetV1 · ImageNettop-1 accuracy: 66.45 -> 71.09Improved
replace ReLU6 clipping with BN-based clipping (8b fixed point) (ablation)MobileNetV1 FX8 · ImageNettop-1 accuracy: 71.60 -> 71.86Improved
replace ReLU6 clipping with BN-based clipping (ablation)DBQ-2T · ImageNettop-1 accuracy: 70.25 -> 70.80Improved
quantize first layer, depthwise layers, FC and activations to 8b (ablation)DBQ-2T MobileNetV1 · ImageNettop-1 accuracy: 70.80 -> 70.92No clear effect
ternarize pointwise layers with one branch (ablation)MobileNetV1 · ImageNettop-1 accuracy: 72.12 -> 66.45Hurt
use two ternary branches instead of two binary branchesResNet-20 DBQ vs LQNet · CIFAR-10accuracy: 90.10 -> 91.06Improved
use two ternary branches instead of three binary branchesResNet-20 DBQ-2T vs LQNet-3B · CIFAR-10accuracy: 92.00 -> 91.93No clear effect
compare DBQ-2T against IAO 8b quantizationMobileNetV1 · ImageNettop-1 accuracy: 69.00 -> 70.92Improved
compare DBQ-2T against RQ 5b quantizationMobileNetV1 · ImageNettop-1 accuracy: 61.50 -> 70.92Improved
apply 2T quantization of PW layers to MobileNetV2MobileNetV2 · ImageNettop-1 accuracy: 71.88 -> 70.54Hurt
apply 2T quantization of PW layers to ShuffleNetV2ShuffleNetV2 · ImageNettop-1 accuracy: 69.36 -> 66.74Hurt

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