Papers › Improving Visual Representation Learning through Perceptual Understanding

Improving Visual Representation Learning through Perceptual Understanding

30 Dec 2022CVPR 2023 1arXiv:2212.14504archive 2025-07-28

Samyakh Tukra, Frederick Hoffman, Ken Chatfield

We present an extension to masked autoencoders (MAE) which improves on the representations learnt by the model by explicitly encouraging the learning of higher scene-level features. We do this by: (i) the introduction of a perceptual similarity term between generated and real images (ii) incorporating several techniques from the adversarial training literature including multi-scale training and adaptive discriminator augmentation. The combination of these results in not only better pixel reconstruction but also representations which appear to capture better higher-level details within images. More consequentially, we show how our method, Perceptual MAE, leads to better performance when used for downstream tasks outperforming previous methods. We achieve 78.1% top-1 accuracy linear probing on ImageNet-1K and up to 88.1% when fine-tuning, with similar results for other downstream tasks, all without use of additional pre-trained models or data.

PaperPDFConference PDFCode

Code

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Representation LearningSelf-Supervised Image Classification

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Self-Supervised Image Classification ImageNet PercMAE (ViT-B, dVAE) Number of Params 80M #30 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet PercMAE (ViT-B, dVAE) Top 1 Accuracy 79.8% #30 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet PercMAE (ViT-B) Number of Params 80M #48 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet PercMAE (ViT-B) Top 1 Accuracy 78.1% #48 of 144 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) PercMAE (ViT-L, dVAE) Number of Params 307M #2 of 65 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) PercMAE (ViT-L, dVAE) Top 1 Accuracy 88.6% #2 of 65 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) PercMAE (ViT-L) Number of Params 307M #5 of 65 Archive leaderboard report
Self-Supervised Image Classification ImageNet (finetuned) PercMAE (ViT-L) Top 1 Accuracy 88.1% #5 of 65 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

MAE

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections