> For the complete documentation index, see [llms.txt](https://docs.biodock.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.biodock.ai/projects-and-models/labeling-tools.md).

# Labeling tools

*What you'll get: how to outline objects with the **Pen**, **AI Select**, **AI Detect** and **AI Prelabel**, how to fix and select labels, and when a tile counts for training.*

These tools work the same way in a project's labeling screen and on the **Label** tab of an AI-assisted analysis. **AI Prelabel** is only in projects, because it uses a model you trained.

{% hint style="info" %}
Labeling needs a desktop or laptop. AI labeling is up to 5× faster in Chrome or Edge than in Safari. Every key is listed in [Labeling shortcuts](https://docs.biodock.ai/projects-and-models/labeling-shortcuts.md).
{% endhint %}

## Pick a class and a tool

1. In the left sidebar, click a class, or press its number (**1**–**9**). The class row now reads **Stop**.
2. Under it, pick a tool: **AI Detect**, **AI Select** or **Pen**. Or press **D**, **S** or **P**. **AI Detect** is only offered for object classes, not region classes.
3. You are now in drawing mode. Hold **Space** and drag to pan. Hold **Shift** to hide every outline for a moment.
4. Press **Esc** to stop drawing.

Which tool to use:

* **AI Detect** is fastest: show it one example and it outlines the look-alikes in the tile.
* **AI Select** outlines one object per click. Use it to fill the gaps.
* **Pen** is slowest but always works. Use it where the AI tools struggle, such as objects cut by the tile edge.
* Once you have trained a model, **AI Prelabel** usually beats all three: your model labels the tile and you only correct it.

## Pen

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-0341ebc2321f2b512193a67edd33926663cde3bf%2Fdocs-labeling-pen.gif?alt=media" alt="Click Nucleus, click Pen, click around the edge of a nucleus point by point and close the shape on the first point"><figcaption><p>Pick the class, click <strong>Pen</strong>, then trace the edge and close the shape.</p></figcaption></figure>

1. Pick the class and click **Pen** (or press **P**).
2. Click around the edge of the object, point by point, and click the first point again to close the shape. Or hold the mouse button and trace the edge in one drag.
3. Made a mistake halfway? Press **Esc**. The half-drawn shape goes and the Pen stays in your hand.

Follow the real border closely. A precise outline teaches the model more than several rough ones.

## AI Select

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-c111a0432bab6f96074b39382f4e02219fae0d3c%2Fdocs-labeling-ai-select.gif?alt=media" alt="Click AI Select, click the tile to prepare it, click inside a nucleus, press A, click the next nucleus and press A"><figcaption><p>Prepare the tile, click inside an object, then press <strong>A</strong> to keep it.</p></figcaption></figure>

1. Pick the class and click **AI Select** (or press **S**).
2. Hover over a tile. It reads "Click tile to prep for AI labeling". Click it once. It reads "Ready!" after a few seconds. If it says it failed to prepare, click it again.
3. Click inside an object. The outline appears.
4. Press **A** to keep it, then click the next object.

To refine an outline before you keep it:

* **Grow it:** click outside the outline where it is missing (a green dot).
* **Shrink it:** click inside the part you don't want (a red dot).
* **Take a dot back:** hold **Cmd** (Mac) or **Ctrl** (Windows) and click it.
* **Touching objects:** press **Esc**, then drag a tight box around just the object you want.

AI Select works inside one tile until you keep or cancel. If you move to another tile first, Biodock asks you to "Finish (A) or cancel (Esc) to label this tile".

## AI Detect

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-b3e8051f7be826abe05fc47edfb9cdeae81a6346%2Fdocs-labeling-ai-detect.gif?alt=media" alt="Click AI Detect, drag a tight box around one typical nucleus, outlines appear across the tile, move the Minimum confidence slider, click Accept"><figcaption><p>Box one example, tune the sliders, then click <strong>Accept</strong>.</p></figcaption></figure>

1. Pick an object class and click **AI Detect** (or press **D**).
2. Drag a tight box around one typical object: not the biggest or the brightest. The tile reads "Success! Now, update the filters & confirm!" and outlines appear on similar objects. Objects you already labeled are skipped.
3. Tune the guesses with the sliders:
   * **Minimum confidence**: higher keeps only the guesses the AI is sure of.
   * **Color similarity**: higher keeps only objects coloured like your example.
   * **Size range**: the left handle drops small objects, the right handle drops large ones.
4. To throw out one wrong guess, click the **✕** in its middle.
5. Click **Accept** (or press **A**) to turn the remaining guesses into labels. **Cancel** (or **Esc**) throws them all away.

**Save filters** keeps the slider settings for this class.

To search the whole image instead of one tile, draw your example box, then click **Apply to all tiles**. The panel shows "Processing tile 1 of …" while it runs. Click **Stop** to end the run early and keep what it has found.

AI Detect uses credits for each tile it runs on. See [Credits, storage and limits](https://docs.biodock.ai/account/credits-storage-and-limits.md).

## AI Prelabel

Once a project has a trained model, it can label tiles for you.

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-de9702a6417ed654edf5f41b62af3a56326d9afa%2Fdocs-labeling-ai-prelabel.gif?alt=media" alt="Click AI Prelabel on the size group card, check the Using model line, click an unlabeled tile, the tile fills with labels and is marked Needs correction"><figcaption><p>Click <strong>AI Prelabel</strong>, then click a tile. It is marked <strong>Needs correction</strong>.</p></figcaption></figure>

1. On the size group's card in the sidebar, click **AI Prelabel**. The **AI Prelabeling Mode** panel opens.
2. Check the line "Using model Version …". Click it to change the **Trained version** or to **Select classes**.
3. Click an unlabeled tile. It reads "Click to AI prelabel" when you hover. A moment later the tile is labeled, its border turns purple, and it is marked **Needs correction**.
4. Click **Stop prelabeling** when you are done.

Then correct each prelabeled tile with the tools on this page and tick **Corrected** in the tile's corner. A prelabeled tile is not used for training until you tick **Corrected**. You can't prelabel a tile that is already **Fully labeled**.

AI Prelabel uses credits. See [Credits, storage and limits](https://docs.biodock.ai/account/credits-storage-and-limits.md).

## Select and fix labels

Press **Esc** to leave drawing mode, then click a label on the image. It is highlighted in the **Labeled objects** list at the bottom of the sidebar.

Each object's row in **Labeled objects** has a settings menu with:

* **Edit**: reshape the outline. Choose **Add** and draw over a missed edge, or **Minus** and draw over a part that spills out. You can also press **E**, then **A** or **M**.
* **Split**: cut one outline into two.
* **Change class**: move the object to another class.
* **Delete**: remove it. Biodock asks before it deletes.

To select several labels, hold **Cmd** (Mac) or **Ctrl** (Windows) and click them, or hold the key and drag across them. The list then shows how many are selected. Click **Actions** to **Change class**, **Merge objects** or **Delete** them all.

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-e0f5e3b9f317e31c4e40641ef6aee03046b72b9e%2Fdocs-labeling-labeled-objects-actions.png?alt=media" alt="Labeled objects panel with several objects selected and the Actions menu open"><figcaption><p>With several labels selected, <strong>Actions</strong> offers <strong>Change class</strong>, <strong>Merge objects</strong> and <strong>Delete</strong>.</p></figcaption></figure>

There is no undo and no save button. Labels save as you go.

## Fully labeled tiles

Only tiles marked **Fully labeled** are used for training.

1. Label every object in the tile, including objects cut off by the tile edge. When the model runs for real, Biodock joins objects across tile edges, so do not change the tile size to avoid cut objects.
2. Hover over the tile. Its corner shows the tile's position and label count.
3. Tick **Fully labeled**. The tile border turns green.

<figure><img src="https://3806122466-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fk3DfNQKoMy1JXdaDLvpG%2Fuploads%2Fgit-blob-1f3f03a7d71e693fb62527a95fbc1c4da6d47883%2Fdocs-labeling-fully-labeled-tile.png?alt=media" alt="A tile corner with the label count and the Fully labeled tick box ticked, and a green tile border"><figcaption><p>Tick <strong>Fully labeled</strong> only when every object in the tile is outlined.</p></figcaption></figure>

Tick it only when the tile is really done. An object you missed in a fully labeled tile teaches the model that it is background.

**Clear** in the same corner removes every label in the tile. It asks first and can't be undone. **Clear** is hidden while the tile is marked **Fully labeled**: untick it first.

The **Tile Borders** key under the image names the border colours: **Fully labeled**, **Prelabeled, needs correction**, **In progress** and **Unlabeled**.

## Next steps

* [Train a model](https://docs.biodock.ai/projects-and-models/training-a-model.md) from your fully labeled tiles.
* [Labeling shortcuts](https://docs.biodock.ai/projects-and-models/labeling-shortcuts.md): every key in the labeling screen.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.biodock.ai/projects-and-models/labeling-tools.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
