# Pre-encoded Mode

The **Pre-encoded Mode** uses pre-encoded image features generated by the
{doc}`encoding` tool. This is ideal when:

- You segment the **same image repeatedly** across multiple sessions.
- You encode on a **remote server** (Colab, AWS) and segment locally.
- You want the **fastest possible prompt response** (no background encoding
  delay).

## Switching to Pre-encoded Mode

1. Open the **Geo-SAM Segmentation** tool.
2. In the **Input / Output** tab, change the source selector dropdown from
    **Live Encoding** to **Pre-encoded**.
3. The input controls switch to show a file picker and a **Load** button.

## Loading a Feature Folder

1. Click the file picker next to **Pre-encoded** and select the folder that
   contains the encoded features (the output directory from the Image
   Encoder, which contains a `features/` subfolder and a `manifest.parquet`).
2. Click **Load**.

```{dropdown} Feature folder structure
:color: info
:open:

A valid feature folder looks like this:

    my_feature_folder/
    ├── manifest.parquet       # chip metadata (bounds, CRS, model info)
    └── features/
        ├── chip_000000.pt     # encoded feature for chip 0
        ├── chip_000001.pt     # encoded feature for chip 1
        └── ...
```

After loading, Geo-SAM displays a success message with the chip count and
model information. The model dropdown is locked to the model that was used
during encoding (recorded in the manifest).

```{admonition} Model must be downloaded
:class: warning

The model used during encoding must be downloaded in Settings > Model
Management. If it is missing, Geo-SAM will prompt you to download it.
```

## Adding the Raster Image

Feature files contain encoded features but not the image itself. **Add the
corresponding raster image to your QGIS project** so you can see what you are
segmenting. The image must cover the same area as the encoded features.

## Zoom to Extent

Click **Zoom to** to navigate the map canvas to the feature folder's extent.
This is useful when the feature folder covers a different area than your
current view.

## Segmentation Workflow

Once the feature folder is loaded, the segmentation workflow is identical to
the Live Encoding mode:

1. Add **foreground points**, **background points**, and **bounding boxes**
   in the **Prompts** tab.
2. View results instantly (no background encoding -- features are already
   available).
3. Save with **S** or the **Save** button.
4. Clear with **C** and move to the next object.

See {doc}`live_encoding_segmentation` for the full reference on prompts, styles,
options, shortcuts, and preview mode -- they are shared between both modes.


## Generating Feature Files

Feature files are created by the **Geo-SAM Image Encoder** tool. See
{doc}`encoding` for the full encoder reference, or {doc}`copilot` for the
interactive parameter tuning tool.