AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

Image-Free Classifier Injection for Zero-Shot Classification

Anders Christensen, Massimiliano Mancini, A. Sophia Koepke, Ole Winther, Zeynep Akata

ICCV 2023 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Generative feature synthesis for zero-shot · 23 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
replace L2 loss with cosine loss in descriptor-to-classifier MLP (ablation)ICIS MLP base model · CUB GZSLharmonic mean H: 0.0 -> 49.5Improved
replace L2 loss with cosine loss (ablation)ICIS MLP base model · SUN ZSLzero-shot accuracy: 49.7 -> 48.5Hurt
add within-space reconstruction losses on descriptors and classifiers (ablation)ICIS · AWA2 ZSLzero-shot accuracy: 50.9 -> 63.5Improved
add within-space reconstruction losses (ablation)ICIS · SUN GZSLharmonic mean H: 29.2 -> 35.4Improved
add across-space reconstruction (weights to descriptors) (ablation)ICIS · CUB ZSLzero-shot accuracy: 58.0 -> 60.1Improved
include unseen class descriptors during training (ablation)ICIS · CUB ZSLzero-shot accuracy: 60.1 -> 60.6No clear effect
include unseen descriptors during training (ablation)ICIS · SUN GZSLharmonic mean H: 33.9 -> 32.7Hurt
use ICIS instead of best baseline on image-free ZSLICIS · CUB, AWA2, SUN ZSLzero-shot accuracy: +15.5% CUB and +9.2% AWA2 over SMO, +7.4% over ConSE on SUNImproved
train ICIS from generic ImageNet classifier instead of seen-class classifierICIS · CUB and SUN ZSL (CLIP embeddings)zero-shot accuracy: -22.5% CUB, -13.7% SUNHurt
train ICIS from ImageNet classifier with 1000 descriptor-classifier pairsICIS · AWA2 ZSL (CLIP embeddings)zero-shot accuracy: up to 86.1Improved
use non-visual embeddings (Wiki2Vec, ConceptNet) with ImageNet-pretrained classifier on CUBICIS · CUB ZSL from ImageNetzero-shot accuracy: 6.0 (wiki2vec), 10.2 (ConceptNet)Hurt

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