> For the complete documentation index, see [llms.txt](https://docs.roboflow.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roboflow.com/models/supported-tasks/action-recognition.md).

# Action Recognition

Find which actions happen in a video and when, then train and deploy your own action recognition model on Roboflow.

## What is action recognition?

Action recognition finds which actions happen in a video and when. The model looks at a window of frames at a time and returns a timeline: each entry has an action class, a start and end frame, and a confidence score. Several actions can overlap. Use it when the answer depends on motion over time, which a single image cannot show.

<figure><img src="https://3448603185-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fcmo9mhfIjYrvFFy1U7dk%2Fuploads%2Fgit-blob-c85273bca818e4b15c38a6f43b3baeabae657d9f%2Fimg-task-action-recognition-wide.gif?alt=media" alt="A knife cutting an orange in half, labeled CUTTING ORANGE while the action happens, with a timeline bar"><figcaption></figcaption></figure>

## Supported action recognition models

Roboflow supports these action recognition models. See the [Action Recognition models](/models/supported-models.md#action-recognition) table for export and deployment support per model.

<table data-search="false"><thead><tr><th>Model</th><th>License</th><th data-type="checkbox">Trainable on Roboflow</th><th>Latest release</th></tr></thead><tbody><tr><td><a href="/models/supported-models/cosmos-3-edge.md">Cosmos 3 Edge</a></td><td><a href="/models/model-licenses.md#other-models">Upstream license</a></td><td>true</td><td>2026</td></tr><tr><td><a href="/models/supported-models/v-jepa-2-1.md">V-JEPA 2.1 (Beta)</a></td><td><a href="/models/model-licenses.md#other-models">Upstream license</a></td><td>true</td><td>2026</td></tr></tbody></table>

## Fine-tuning an action recognition model

{% stepper %}
{% step %}

### Create an action recognition project

Create a [Project](https://docs.roboflow.com/datasets/create-and-upload/create-a-project) and pick "Action Recognition" as the project type.

<figure><img src="https://3448603185-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fcmo9mhfIjYrvFFy1U7dk%2Fuploads%2Fgit-blob-d8330839d4ddd189368ee981bb1af48717845d8f%2Fcreate-project-action-recognition.webp?alt=media" alt="Create project form with Action Recognition selected as the project type and a preview of an action recognition example"><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Upload and annotate videos

[Upload videos](https://docs.roboflow.com/datasets/create-and-upload/adding-data), then [label video segments](https://docs.roboflow.com/datasets/annotate/annotate/annotate-video-segments) on the timeline in the video editor: each segment marks when an action class starts and ends. You label segments by hand. Put videos in both the train and valid splits.

<figure><img src="https://3448603185-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fcmo9mhfIjYrvFFy1U7dk%2Fuploads%2Fgit-blob-9fd16caace74692b63381cfad87666ff54b81715%2Fannotate-action-recognition.webp?alt=media" alt="Video annotation editor with turn, back salto, front salto, and leap jump hop segments on separate timeline rows below a gymnastics video"><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Generate a dataset version

[Generate a dataset version](https://docs.roboflow.com/datasets/versions/dataset-versions/create-a-dataset-version): a frozen snapshot of your labeled videos and their splits.
{% endstep %}

{% step %}

### Train an action recognition model

Click "Train" and pick Cosmos 3 Edge or V-JEPA 2.1. The [V-JEPA 2.1](/models/supported-models/v-jepa-2-1.md) page lists the recommended video sampling. See [Train a Model](/models/train/train-a-model.md) for every option.

<figure><img src="https://3448603185-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fcmo9mhfIjYrvFFy1U7dk%2Fuploads%2Fgit-blob-a1c85d104e3719a171ae36b429ab2554fd82b319%2Ftrain-architectures-action-recognition.webp?alt=media" alt="Select Architecture step on the Train page listing V-JEPA 2.1 and Cosmos 3"><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

## Deploy an action recognition model

After training, run your action recognition model on video files or live streams with the [Serverless Cloud API](https://docs.roboflow.com/deployment/roboflow-cloud/serverless-api) or on your own hardware with [Roboflow Inference](https://docs.roboflow.com/deployment/self-hosted/self-hosted). See [Choosing a Deployment](https://docs.roboflow.com/deployment/choosing-a-deployment) to compare the options.