AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Demystifying Randomly Initialized Networks for Evaluating Generative Models

Junghyuk Lee, Jun-Hyuk Kim, Jong-Seok Lee

AAAI 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Data augmentation and generalization bias · 1 citations (OpenAlex)

Read the paper

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
Use random Swin features rather than trained SwAV features for FID evaluationGenerative models across ImageNet, FFHQ, and Pokémon · ImageNet 256², ImageNet 64², FFHQ 1024²/256², Pokémon 256²FID model rankingMixed
Use random features to assess low-level image differences such as color, blur, and noiseFID evaluation of disturbed ImageNet images · ImageNet validationFID sensitivity to low-level disturbance: Gaussian noise level 3: Inception (trained) 41.09; Swin (random) 0.16; color jitter level 3: Inception (trained) 28.16; Improved
Use trained features to assess high-level semantic changes rather than random features aloneFID evaluation of outlier-replaced FFHQ sets · FFHQFID sensitivity to high-level outliersMixed
Combine trained and random features to capture distinct types of generated-image abnormalitiesGenerative-model evaluation · FFHQ and ImageNetcoverage of high- and low-level outliers: 0.276, 0.238, 0.152; 0.191 and 0.127Improved
Use Transformer rather than CNN random featuresRandom-feature FID evaluation · Generative models across datasetsvariation across random seedsImproved
Use Swin-T or ViT-T instead of larger tested models (ablation)Random-feature FID evaluation · ImageNet 256²model ranking consistency: BigGAN: Swin-T 1.07, Swin-S 1.07, Swin-B 1.54; ViT-T 0.62, ViT-B 3.14No clear effect
Increase Swin size from T to S (ablation)Random-feature evaluation of BigGAN and ADM-U · ImageNet 256²FID scale: BigGAN: 1.07 -> 1.07; ADM-U: 3.57 -> 3.70No clear effect
Increase Swin size from T to B (ablation)Random-feature evaluation of BigGAN and ADM-U · ImageNet 256²FID scale: BigGAN: 1.07 -> 1.54; ADM-U: 3.57 -> 5.03Mixed
Increase ViT size from T to B (ablation)Random-feature evaluation of BigGAN and ADM-U · ImageNet 256²FID scale: BigGAN: 0.62 -> 3.14; ADM-U: 1.61 -> 6.61Mixed
Average random-feature results across different initialization seedsRandom-feature FID evaluation · Generative models across datasetsvariation across seedsImproved

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