AI Paper Atlas

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

Quality Diversity for Visual Pre-Training

Ruchika Chavhan, Henry Gouk, Da Li, Timothy Hospedales

ICCV 2023 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 3 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
pre-train an ensemble with diversity loss over augmentation invariances plus KL to a baselinesupervised ResNet50 on IN1K · 14 downstream tasks, linear readoutaverage rank: 2.7 -> 1.4Improved
QD pre-training versus supervised baseline on spatially sensitive regressionsupervised ResNet50 · CelebA and Causal3DR2 (%): 49.0 -> 61.5 (CelebA); 64.1 -> 72.8 (Causal3D)Improved
QD pre-training versus baseline on CIFAR10supervised ResNet50 · CIFAR10accuracy (%): 90.4 -> 90.3No clear effect
use multihead ensemble with quality loss only instead of a diversity loss (ablation)supervised ResNet50 · 14 downstream tasksaverage rank: 2.7 -> 3.7 (Baseline vs Multihead)Hurt
fuse ensemble with learned stacking weights instead of concatenation (ablation)QD4V ensemble · Animal Pose (supervised)R2 (%): 11.2 -> 12.5Improved
fuse ensemble with learned stacking weights instead of unweighted averaging (ablation)QD4V ensemble · ALOI (MoCo IN1K)R2 (%): 42.9 -> 47.4Improved
QD pre-training on few-shot regressionsupervised ResNet50 · 300W s=0.05R2 (%): 20.5 -> 37.5Improved
QD pre-training on few-shot Flowers (5,1) versus diverse ensemblesupervised ResNet50 · Flowers (5,1)accuracy (%): 77.3 -> 73.4 vs baselineHurt
QD pre-training with frozen backbone for dense taskssupervised ResNet50 · Cityscapes and ADE20kmIoU: 39.6 -> 41.0; 30.4 -> 33.8Improved
QD pre-training for MoCo detection on Pascal VOCMoCo ResNet50 · Pascal VOC detectionAP: 54.2 -> 53.3Hurt
scale pre-training data (CLIP) instead of choosing invariancesCLIP RN50 versus QD4V · pose-sensitive tasksdownstream performanceHurt

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