Browse › Computer Vision › Action Recognition › AVA v2.2
AVA v2.2 Benchmark (Action Recognition)
Action Recognition is a computer vision task that involves recognizing human actions in videos or images. The goal is to classify and categorize the actions being performed in the video or image into a predefined set of action classes.
In the video domain, it is an open question whether training an action classification network on a sufficiently large dataset, will give a similar boost in performance when applied to a different temporal task or dataset. The challenges of building video datasets has meant that most popular benchmarks for action recognition are small, having on the order of 10k videos.
Please note some benchmarks may be located in the Action Classification or Video Classification tasks, e.g. Kinetics-400.
The archive carries no text for this table; the description above is the archive's text for the task Action Recognition. archive 2025-07-28
Over time archive 2025-07-28
The chart needs JavaScript; the table below carries every value.
Direction inferred from the metric name, not from the archive: mAP (higher is better). Points are placed at the row's paper date; 38 of 38 rows carry one.
Results archive 2025-07-28
Archive rows end at the archive snapshot, 2025-07-28: no result published after that date is in this table. Rank is the archive's row order at that snapshot; not re-ranked here. Metric values are the archive's strings. Column headers sort the table in your browser; each row keeps its archive rank.
| Paper | Code | Ran Syntology | Report | |||||
|---|---|---|---|---|---|---|---|---|
| 1 | LART (Hiera-H, K700 PT+FT) | 45.1 | ✓ | Paper | Code | 2023 | linked, not harvested | report |
| 2 | Hiera-H (K700 PT+FT) | 43.3 | ✓ | Paper | Code | 2023 | 0 of 6 ran · 6 unverified | report |
| 3 | VideoMAE V2-g | 42.6 | ✓ | Paper | Code | 2023 | 2 of 6 ran · 4 unverified | report |
| 4 | STAR/L | 41.7 | ✓ | Paper | – | 2023 | no code linked | report |
| 5 | MVD (Kinetics400 pretrain+finetune, ViT-H, 16x4) | 41.1 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 6 | InternVideo | 41.01 | ✓ | Paper | Code | 2022 | 3 of 3 ran · 0 unverified | report |
| 7 | MVD (Kinetics400 pretrain, ViT-H, 16x4) | 40.1 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 8 | MaskFeat (Kinetics-600 pretrain, MViT-L) | 39.8 | ✓ | Paper | Code | 2021 | linked, not harvested | report |
| 9 | UMT-L (ViT-L/16) | 39.8 | ✓ | Paper | Code | 2023 | 3 of 8 ran · 5 unverified | report |
| 10 | VideoMAE (K400 pretrain+finetune, ViT-H, 16x4) | 39.5 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 11 | VideoMAE (K700 pretrain+finetune, ViT-L, 16x4) | 39.3 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 12 | MVD (Kinetics400 pretrain+finetune, ViT-L, 16x4) | 38.7 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 13 | VideoMAE (K400 pretrain+finetune, ViT-L, 16x4) | 37.8 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 14 | MVD (Kinetics400 pretrain, ViT-L, 16x4) | 37.7 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 15 | VideoMAE (K400 pretrain, ViT-H, 16x4) | 36.5 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 16 | VideoMAE (K700 pretrain, ViT-L, 16x4) | 36.1 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 17 | MeMViT-24 | 35.4 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 18 | MViTv2-L (IN21k, K700) | 34.4 | ✓ | Paper | Code | 2021 | linked, not harvested | report |
| 19 | VideoMAE (K400 pretrain, ViT-L, 16x4) | 34.3 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 20 | MVD (Kinetics400 pretrain+finetune, ViT-B, 16x4) | 34.2 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 21 | AMD(ViT-B/16) | 33.5 | ✓ | Paper | – | 2023 | no code linked | report |
| 22 | HIT | 32.6 | – | Paper | Code | 2022 | linked, not harvested | report |
| 23 | VideoMAE (K400 pretrain+finetune, ViT-B, 16x4) | 31.8 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 24 | ACAR-Net, SlowFast R-101 (Kinetics-700 pretraining) | 31.72 | ✓ | Paper | Code | 2020 | 1 of 27 ran · 26 unverified | report |
| 25 | MVD (Kinetics400 pretrain, ViT-B, 16x4) | 31.1 | ✓ | Paper | Code | 2022 | linked, not harvested | report |
| 26 | Object Transformer | 31.0 | – | Paper | Code | 2021 | 5 of 19 ran · 14 unverified | report |
| 27 | MViT-B-24, 32x3 (Kinetics-600 pretraining) | 28.7 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 28 | MViT-B, 32x3 (Kinetics-500 pretraining) | 27.5 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 29 | SlowFast, 16x8 R101+NL (Kinetics-600 pretraining) | 27.5 | – | Paper | Code | 2018 | 0 of 10 ran · 10 unverified | report |
| 30 | MViT-B, 64x3 (Kinetics-400 pretraining) | 27.3 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 31 | SlowFast, 8x8 R101+NL (Kinetics-600 pretraining) | 27.1 | – | Paper | Code | 2018 | 0 of 10 ran · 10 unverified | report |
| 32 | MViT-B, 32x3 (Kinetics-400 pretraining) | 26.8 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 33 | VideoMAE (K400 pretrain, ViT-B, 16x4) | 26.7 | ✓ | Paper | Code | 2022 | 9 of 13 ran · 4 unverified | report |
| 34 | ORViT MViT-B, 16x4 (K400 pretraining) | 26.6 | – | Paper | Code | 2021 | 1 of 7 ran · 6 unverified | report |
| 35 | MViT-B, 16x4 (Kinetics-600 pretraining) | 26.1 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 36 | MViT-B, 16x4 (Kinetics-400 pretraining) | 24.5 | – | Paper | Code | 2021 | 13 of 26 ran · 13 unverified | report |
| 37 | SlowFast, 8x8, R101 (Kinetics-400 pretraining) | 23.8 | – | Paper | Code | 2018 | 0 of 10 ran · 10 unverified | report |
| 38 | SlowFast, 4x16, R50 (Kinetics-400 pretraining) | 21.9 | – | Paper | Code | 2018 | 0 of 10 ran · 10 unverified | report |
All 38 rows shown. 38 link to a paper page on this site; 25 are marked as using additional training data in the archive. No GitHub stars are tracked; "Code" is the first repository the archive lists for the row. The archive carries no row tags, review links or community-submitted rows for this table; none are shown. archive 2025-07-28
Syntology Ran reads "N of M ran · U unverified": of the M code samples Syntology harvested from repositories linked to that row's paper (joined by arXiv id), N executed on a synthesized input and the other U = M−N are unverified (harvested, no recorded run). It counts code from repositories linked to that row's paper, not this result: the row's number was not reproduced and nothing here is a correctness claim. The other cell texts mean no graph line for the row: "linked, not harvested" (the archive links code, Syntology has not harvested it), "no code linked" (no code link in the archive), "not matched" (the row's paper URL matched no paper on this site). 25 rows have a graph line, from 10 distinct papers; 20 rows (8 papers) have at least one sample that ran. Counting each paper once: Syntology ran 37 of 125 samples; 88 unverified. Separately, 24 of those 125 are pointer-only (licence): the site points at that code rather than redistributing it, a licence property recorded for ran and unverified samples alike; each cell's tooltip carries the row's own pointer-only count. Read from the graph 2026-09-24. Per-sample status is on the paper page.
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