Browse source images#

The Browse tab is a deep-zoom viewer for the raw input images under your selected source root — before you build a pipeline or run anything. It lists every image under the source folder with two cascading dropdowns (dataset folder → image) and a ‹/› stepper, and renders any one in an OpenSeadragon viewport with a metadata panel.

It is designed to work offline over an SSH tunnel: OpenSeadragon is vendored (no CDN), each source file is lazily normalized to an 8-bit RGB PNG, and the tiles live in an ephemeral temp cache that is wiped each session.

Step 1 - Open the Browse tab#

Browse is a leaf tab immediately after Home in the top bar. Open it, or navigate to /browse/:

The Browse tab with no source root selected, showing the empty hint.

With no source root selected, the page shows a hint asking you to pick one from the top bar. The Browse tab reads the shared source root — the same source: status control the rest of the hub uses — so once a source is set, every page that consumes it (including Browse) updates together.

Step 2 - Select a source#

Click the source: status in the top bar to open the sandbox-bounded directory picker, navigate to the folder holding your images, and confirm. The Browse tab reacts immediately:

The Browse tab with a source root set: dataset + image dropdowns populated and the first plate deep-zoomed in the OpenSeadragon viewport with the metadata panel below.

  • Dataset dropdown — images are grouped by their subfolder relative to the source root (. is shown as (root)). When the source folder is flat (all images directly under it), the dataset dropdown is hidden and you only see the image picker.

  • Image dropdown — lists the files in the selected dataset. The first image auto-selects so the viewport is never blank.

  • ‹ / › stepper — moves to the previous / next image within the current dataset. The buttons disable at the first and last image so stepping never wraps around.

Step 3 - Browse and zoom#

The viewport is OpenSeadragon, so zoom (scroll / pinch) and pan (drag) are GPU-smooth even on large plate scans. Each source file is tiled into a deep-zoom (DZI) image pyramid on first view, so only the tiles you are looking at are loaded — large RAW or TIFF scans open responsively.

Images are rendered faithfully: any supported format (standard formats and camera RAW alike) is decoded through phenotypic.Image and downcast to 8-bit with a full-range conversion — no auto-contrast or histogram stretching — so what you see matches the pixel data the pipeline will operate on.

Step 4 - Read the metadata#

Below the viewport, the metadata panel reports, for the current image:

Field

Source

Dimensions

Pixel width × height.

Size

On-disk file size (human-readable).

Captured

EXIF capture timestamp, when present.

Camera

EXIF camera make + model, when present.

EXIF is pulled from the image’s imported metadata (populated by exifread for JPEG and TIFF-based RAW such as NEF / CR2). Any field that is absent or unreadable is omitted — the panel degrades gracefully rather than erroring, so a plain PNG simply shows — for the EXIF fields.

Note

The tile cache is ephemeral. Normalized PNGs and their DZI tiles are written under tempfile.gettempdir()/phenotypic/browse, keyed by the image’s path. The cache is wiped on launch and again at process exit, so nothing persists between sessions and the cache never accumulates stale tiles. RAW that cannot be decoded on the current platform (e.g. camera RAW on Windows, where rawpy is unavailable) surfaces an inline viewer notice instead of a broken tile.

Where to next#

  • Build a Pipeline — once you have eyeballed the input images, compose the pipeline that will process them.

  • View Results — after a run, the Results viewer renders each plate with detection overlays and the measurements table.