Papers › Asymmetric Masked Distillation for Pre-Training Small Foundation Models

Asymmetric Masked Distillation for Pre-Training Small Foundation Models

6 Nov 2023CVPR 2024 1arXiv:2311.03149archive 2025-07-28

Zhiyu Zhao, Bingkun Huang, Sen Xing, Gangshan Wu, Yu Qiao, LiMin Wang

Self-supervised foundation models have shown great potential in computer vision thanks to the pre-training paradigm of masked autoencoding. Scale is a primary factor influencing the performance of these foundation models. However, these large foundation models often result in high computational cost. This paper focuses on pre-training relatively small vision transformer models that could be efficiently adapted to downstream tasks. Specifically, taking inspiration from knowledge distillation in model compression, we propose a new asymmetric masked distillation (AMD) framework for pre-training relatively small models with autoencoding. The core of AMD is to devise an asymmetric masking strategy, where the teacher model is enabled to see more context information with a lower masking ratio, while the student model is still equipped with a high masking ratio. We design customized multi-layer feature alignment between the teacher encoder and student encoder to regularize the pre-training of student MAE. To demonstrate the effectiveness and versatility of AMD, we apply it to both ImageMAE and VideoMAE for pre-training relatively small ViT models. AMD achieved 84.6% classification accuracy on IN1K using the ViT-B model. And AMD achieves 73.3% classification accuracy using the ViT-B model on the Something-in-Something V2 dataset, a 3.7% improvement over the original ViT-B model from VideoMAE. We also transfer AMD pre-trained models to downstream tasks and obtain consistent performance improvement over the original masked autoencoding. The code and models are available at https://github.com/MCG-NJU/AMD.

PaperPDFConference PDF

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

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

Tasks

Action ClassificationAction RecognitionImage ClassificationKnowledge DistillationModel Compression

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Action Classification Kinetics-400 AMD(ViT-B/16) Acc@1 82.2 #79 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-B/16) Acc@5 95.3 #79 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-B/16) FLOPs (G) x views 180x15 #79 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-B/16) Parameters (M) 87 #79 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-S/16) Acc@1 80.1 #104 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-S/16) Acc@5 94.5 #104 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-S/16) FLOPs (G) x views 57X15 #104 of 207 Archive leaderboard report
Action Classification Kinetics-400 AMD(ViT-S/16) Parameters (M) 22 #104 of 207 Archive leaderboard report
Action Recognition AVA v2.2 AMD(ViT-B/16) mAP 33.5 #21 of 38 Archive leaderboard report
Action Recognition HMDB-51 AMD(ViT-B/16) Average accuracy of 3 splits 79.6 #24 of 77 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-B/16) GFLOPs 180x6 #22 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-B/16) Parameters 87 #22 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-B/16) Top-1 Accuracy 73.3 #22 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-B/16) Top-5 Accuracy 94.0 #22 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-S/16) GFLOPs 57x6 #37 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-S/16) Parameters 22 #37 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-S/16) Top-1 Accuracy 70.2 #37 of 123 Archive leaderboard report
Action Recognition Something-Something V2 AMD(ViT-S/16) Top-5 Accuracy 92.5 #37 of 123 Archive leaderboard report
Action Recognition UCF101 AMD(ViT-B/16) 3-fold Accuracy 97.1 #25 of 91 Archive leaderboard report
Image Classification ImageNet AMD(ViT-B/16) Number of params 87M #305 of 1060 Archive leaderboard report
Image Classification ImageNet AMD(ViT-B/16) Top 1 Accuracy 84.6% #305 of 1060 Archive leaderboard report
Image Classification ImageNet AMD(ViT-S/16) Number of params 22M #573 of 1060 Archive leaderboard report
Image Classification ImageNet AMD(ViT-S/16) Top 1 Accuracy 82.1% #573 of 1060 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

AttentionDense ConnectionsKnowledge DistillationLayer NormalizationLinear LayerMAEMulti-Head AttentionResidual ConnectionSoftmaxVision Transformer

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