AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

MultiSiam: Self-Supervised Multi-Instance Siamese Representation Learning for Autonomous Driving

Kai Chen, Lanqing Hong, Hang Xu, Zhenguo Li, Dit-Yan Yeung

ICCV 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Pretext-task visual pre-training · 39 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 IoU threshold 0.5 to random cropping (ablation)BYOL · Cityscapes val mIoU (ablation)mIoU: 67.2 -> 69.2Improved
add intra-image clustering loss (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 69.2 -> 70.0Improved
add self-attention to the predictor (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 70.0 -> 71.2Improved
add offset alignment (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 71.2 -> 71.9Improved
add feature alignment (RoI) to 2D feature maps (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 68.8 -> 70.0Improved
normalize offset coordinates (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 71.2 -> 71.9Improved
raise IoU threshold from 0.5 to 0.7 (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 70.0 -> 69.2Hurt
use K=3 clusters instead of K=4 (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 68.4 -> 70.0Improved
predict cluster centroids instead of dense clustering (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 69.0 -> 70.0Improved
use K-means cluster targets instead of per-pixel targets (ablation)MultiSiam · CityscapesmIoU: 71.0 -> 74.1Improved
add residual connection in self-attention with offset alignment (ablation)MultiSiam · Cityscapes val mIoU (ablation)mIoU: 71.9 -> 70.0Hurt
swap pre-training data from ImageNet to WaymoMoCo-v2 · Cityscapes semantic segmentationmIoU: 75.7 -> 73.6Hurt

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