{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/asymmetric-masked-distillation-for-pre","title":"Asymmetric Masked Distillation for Pre-Training Small Foundation Models","arxiv_id":"2311.03149","date":"2023-11-06","proceeding":"CVPR 2024 1","authors":["Zhiyu Zhao","Bingkun Huang","Sen Xing","Gangshan Wu","Yu Qiao","LiMin Wang"],"abstract":"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.","url_abs":"https://arxiv.org/abs/2311.03149v2","url_pdf":"https://arxiv.org/pdf/2311.03149v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[],"tasks":[{"task_slug":"action-classification","task_name":"Action Classification"},{"task_slug":"action-recognition-in-videos","task_name":"Action Recognition"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"knowledge-distillation","task_name":"Knowledge Distillation"},{"task_slug":"model-compression","task_name":"Model Compression"}],"methods":[{"method_slug":"attention","method_name":"Attention"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"knowledge-distillation","method_name":"Knowledge Distillation"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"mae","method_name":"MAE"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"vision-transformer","method_name":"Vision Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-classification-on-kinetics-400","task":"Action Classification","dataset":"Kinetics-400","model":"AMD(ViT-B/16)","rank_in_archive_order":79,"of":207,"metrics":{"Acc@1":"82.2","Acc@5":"95.3","FLOPs (G) x views":"180x15","Parameters (M)":"87"},"uses_additional_data":false},{"leaderboard":"/sota/action-classification-on-kinetics-400","task":"Action Classification","dataset":"Kinetics-400","model":"AMD(ViT-S/16)","rank_in_archive_order":104,"of":207,"metrics":{"Acc@1":"80.1","Acc@5":"94.5","FLOPs (G) x views":"57X15","Parameters (M)":"22"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-on-ava-v2-2","task":"Action Recognition","dataset":"AVA v2.2","model":"AMD(ViT-B/16)","rank_in_archive_order":21,"of":38,"metrics":{"mAP":"33.5"},"uses_additional_data":true},{"leaderboard":"/sota/action-recognition-in-videos-on-hmdb-51","task":"Action Recognition","dataset":"HMDB-51","model":"AMD(ViT-B/16)","rank_in_archive_order":24,"of":77,"metrics":{"Average accuracy of 3 splits":"79.6"},"uses_additional_data":true},{"leaderboard":"/sota/action-recognition-in-videos-on-something","task":"Action Recognition","dataset":"Something-Something V2","model":"AMD(ViT-B/16)","rank_in_archive_order":22,"of":123,"metrics":{"GFLOPs":"180x6","Parameters":"87","Top-1 Accuracy":"73.3","Top-5 Accuracy":"94.0"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-something","task":"Action Recognition","dataset":"Something-Something V2","model":"AMD(ViT-S/16)","rank_in_archive_order":37,"of":123,"metrics":{"GFLOPs":"57x6","Parameters":"22","Top-1 Accuracy":"70.2","Top-5 Accuracy":"92.5"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-ucf101","task":"Action Recognition","dataset":"UCF101","model":"AMD(ViT-B/16)","rank_in_archive_order":25,"of":91,"metrics":{"3-fold Accuracy":"97.1"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"AMD(ViT-B/16)","rank_in_archive_order":305,"of":1060,"metrics":{"Number of params":"87M","Top 1 Accuracy":"84.6%"},"uses_additional_data":false},{"leaderboard":"/sota/image-classification-on-imagenet","task":"Image Classification","dataset":"ImageNet","model":"AMD(ViT-S/16)","rank_in_archive_order":573,"of":1060,"metrics":{"Number of params":"22M","Top 1 Accuracy":"82.1%"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2311.03149","atlas_url":"https://app.syntology.ai/?focus=2311.03149","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}