</> Code with Papers by Syntology

Papers › MViTv2: Improved Multiscale Vision Transformers for Classification and Detection

MViTv2: Improved Multiscale Vision Transformers for Classification and Detection

2 Dec 2021CVPR 2022 1arXiv:2112.01526archive 2025-07-28

Yanghao Li, Chao-yuan Wu, Haoqi Fan, Karttikeya Mangalam, Bo Xiong, Jitendra Malik, Christoph Feichtenhofer

In this paper, we study Multiscale Vision Transformers (MViTv2) as a unified architecture for image and video classification, as well as object detection. We present an improved version of MViT that incorporates decomposed relative positional embeddings and residual pooling connections. We instantiate this architecture in five sizes and evaluate it for ImageNet classification, COCO detection and Kinetics video recognition where it outperforms prior work. We further compare MViTv2s' pooling attention to window attention mechanisms where it outperforms the latter in accuracy/compute. Without bells-and-whistles, MViTv2 has state-of-the-art performance in 3 domains: 88.8% accuracy on ImageNet classification, 58.7 boxAP on COCO object detection as well as 86.1% on Kinetics-400 video classification. Code and models are available at https://github.com/facebookresearch/mvit.

PaperPDFConference PDFCode

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

Code

facebookresearch/SlowFast officialmentioned in papermentioned on GitHubpytorchApache-2.0 report
facebookresearch/mvit officialmentioned on GitHubpytorchApache-2.0 report
facebookresearch/detectron2 officialpytorch report
3dperceptionlab/visual-wetlandbirds mentioned on GitHubpytorchNOASSERTION report
JunweiLiang/aicity_action mentioned on GitHubpytorch report
rajatmodi62/occludedactionbenchmark mentioned on GitHubpytorch report
rwightman/pytorch-image-models mentioned on GitHubpytorch report
open-mmlab/mmclassification pytorchApache-2.0 report
gitlab.com/birder/birder pytorch report

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

Action ClassificationAction RecognitionImage ClassificationInstance SegmentationObjectObject DetectionVideo ClassificationVideo Recognition

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Action Classification Kinetics-400 MViTv2-L (ImageNet-21k pretrain) Acc@1 86.1 #51 of 207 Archive leaderboard report
Action Classification Kinetics-400 MViTv2-L (ImageNet-21k pretrain) Acc@5 97.0 #51 of 207 Archive leaderboard report
Action Classification Kinetics-400 MViT-B (train from scratch) FLOPs (G) x views 225x5 #207 of 207 Archive leaderboard report
Action Classification Kinetics-600 MViTv2-L (ImageNet-21k pretrain) Top-1 Accuracy 87.9 #23 of 65 Archive leaderboard report
Action Classification Kinetics-600 MViTv2-L (ImageNet-21k pretrain) Top-5 Accuracy 97.9 #23 of 65 Archive leaderboard report
Action Classification Kinetics-600 MViTv2-L (train from scratch) Top-1 Accuracy 85.5 #30 of 65 Archive leaderboard report
Action Classification Kinetics-600 MViTv2-B (train from scratch) Top-5 Accuracy 97.2 #64 of 65 Archive leaderboard report
Action Classification Kinetics-600 MViT-L (train from scratch) GFLOPs 206x5 #65 of 65 Archive leaderboard report
Action Classification Kinetics-700 MViTv2-L (ImageNet-21k pretrain) Top-1 Accuracy 79.4 #16 of 36 Archive leaderboard report
Action Classification Kinetics-700 MViTv2-L (ImageNet-21k pretrain) Top-5 Accuracy 94.9 #16 of 36 Archive leaderboard report
Action Classification Kinetics-700 MoViNet-A6 Top-1 Accuracy 79.4 #17 of 36 Archive leaderboard report
Action Classification Kinetics-700 MViTv2-B Top-1 Accuracy 76.6 #19 of 36 Archive leaderboard report
Action Classification Kinetics-700 MViTv2-B Top-5 Accuracy 93.2 #19 of 36 Archive leaderboard report
Action Recognition AVA v2.2 MViTv2-L (IN21k, K700) mAP 34.4 #18 of 38 Archive leaderboard report
Action Recognition Something-Something V2 MViTv2-L (IN-21K + Kinetics400 pretrain) Parameters 213.1 #21 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViTv2-L (IN-21K + Kinetics400 pretrain) Top-1 Accuracy 73.3 #21 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViTv2-L (IN-21K + Kinetics400 pretrain) Top-5 Accuracy 94.1 #21 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViT-B (IN-21K + Kinetics400 pretrain) GFLOPs 225x3 #26 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViT-B (IN-21K + Kinetics400 pretrain) Top-1 Accuracy 72.1 #26 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViTv2-B (IN-21K + Kinetics400 pretrain) Parameters 51.1 #120 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViTv2-B (IN-21K + Kinetics400 pretrain) Top-5 Accuracy 93.4 #120 of 123 Archive leaderboard report
Action Recognition Something-Something V2 MViT-L (IN-21K + Kinetics400 pretrain) GFLOPs 2828x3 #123 of 123 Archive leaderboard report
Image Classification ImageNet MViTv2-H (512 res, ImageNet-21k pretrain) GFLOPs 763.5 #31 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-H (512 res, ImageNet-21k pretrain) Number of params 667M #31 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-H (512 res, ImageNet-21k pretrain) Top 1 Accuracy 88.8% #31 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res, ImageNet-21k pretrain) GFLOPs 140.7 #47 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res, ImageNet-21k pretrain) Number of params 218M #47 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res, ImageNet-21k pretrain) Top 1 Accuracy 88.4% #47 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-H (mageNet-21k pretrain) GFLOPs 120.6 #65 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-H (mageNet-21k pretrain) Number of params 667M #65 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-H (mageNet-21k pretrain) Top 1 Accuracy 88% #65 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res) GFLOPs 140.2 #161 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res) Number of params 218M #161 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-L (384 res) Top 1 Accuracy 86.3% #161 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-T GFLOPs 4.7 #550 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-T Number of params 24M #550 of 1060 Archive leaderboard report
Image Classification ImageNet MViTv2-T Top 1 Accuracy 82.3% #550 of 1060 Archive leaderboard report
Instance Segmentation COCO minival MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) mask AP 50.5 #23 of 93 Archive leaderboard report
Instance Segmentation COCO minival MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) mask AP 48.5 #36 of 93 Archive leaderboard report
Instance Segmentation COCO minival MViTv2-L (Cascade Mask R-CNN, single-scale) mask AP 47.1 #39 of 93 Archive leaderboard report
Instance Segmentation COCO minival MViT-L (Mask R-CNN, single-scale) mask AP 46.2 #44 of 93 Archive leaderboard report
Object Detection COCO minival MViTv2-L (Cascade Mask R-CNN, multi-scale, IN21k pre-train) box AP 58.7 #34 of 220 Archive leaderboard report
Object Detection COCO minival MViTv2-H (Cascade Mask R-CNN, single-scale, IN21k pre-train) box AP 56.1 #46 of 220 Archive leaderboard report
Object Detection COCO minival MViTv2-L (Cascade Mask R-CNN, single-scale) box AP 54.3 #57 of 220 Archive leaderboard report
Object Detection COCO minival MViT-L (Mask R-CNN, single-scale, IN21k pre-train) box AP 52.7 #64 of 220 Archive leaderboard report
Object Detection COCO-O MViTV2-H (Cascade Mask R-CNN) Average mAP 30.9 #16 of 45 Archive leaderboard report
Object Detection COCO-O MViTV2-H (Cascade Mask R-CNN) Effective Robustness 5.62 #16 of 45 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

MViT

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