AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Rethinking Pre-training and Self-training

Barret Zoph, Golnaz Ghiasi, Tsung-Yi Lin, Yin Cui, Hanxiao Liu, Ekin Dogus Cubuk, Quoc V. Le

NeurIPS 2020 · Detection and segmentation › Open-set and few-shot detection · approach: Open-set and unknown object detection · 392 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
ImageNet supervised pre-training under strong augmentationRetinaNet EfficientNet-B7 · COCO 100% Augment-S4AP: 44.3 -> 43.3Hurt
ImageNet supervised pre-training under weak augmentationRetinaNet EfficientNet-B7 · COCO 100% Augment-S1AP: 39.2 -> 39.5Improved
add self-training on ImageNet pseudo labelsRetinaNet EfficientNet-B7 Rand Init · COCO all augmentation strengthsAP: +1.3 to +1.7 APImproved
add self-training on top of ImageNet pre-trainingImageNet Init · COCO 20% Augment-S4AP: 33.3 -> 36.0Improved
ImageNet pre-training in low-data regimeRetinaNet EfficientNet-B7 · COCO 20% Augment-S4AP: 30.7 -> 33.3Improved
SimCLR self-supervised pre-trainingResNet-50 RetinaNet · COCO 100% Augment-S4AP: 41.1 -> 40.4Hurt
add self-training with Open ImagesSpineNet-190 · COCO test-devAP: 52.8 -> 54.3Improved
add self-training with PASCAL augEff-L2 NAS-FPN · PASCAL VOC 2012 valmIOU: 88.7 -> 90.0Improved
add noisy human-labeled aug set under strong augmentationEfficientNet-B7 ImageNet Init · PASCAL VOC 2012 val Augment-S4mIOU: 85.2 -> 84.8Hurt
joint training with ImageNet classificationRand Init · COCO 20% Augment-S4AP: 30.7 -> 33.6Improved
use OID instead of ImageNet as unlabeled source (ablation)self-training EfficientNet-B7 · COCO Augment-S4AP: 45.6 -> 46.0Improved
initialize student from teacher instead of random (ablation)self-training EfficientNet-B7 · COCO Augment-S4AP: 45.2 -> 45.6Improved

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