Files
espresso_frame/server/app/face_labels.py
T
tfaour 99069ba5fe
Build and push server image / test (push) Successful in 19s
Build and push server image / build-and-push (push) Successful in 1m58s
Widget system Phase 3: calendar widgets become size-aware
calendar_render.py's _build_* functions now take a real target box
and pick font sizes/margins from three discrete size tiers (nearest
pixel-area fit) instead of always laying out at full panel size and
resizing after the fact -- a calendar widget placed smaller than the
full panel gets an actually-legible layout instead of shrunk text.
Month view falls back to agenda below the smallest tier, where 7
columns can no longer stay readable.

The old photo-inlay split (inlay_region/_content_region/_paste_inlay)
is deleted along with it -- arbitrary widget placement already
subsumes what a fixed half-panel split did, and every call site has
passed photo_inlay=None since the Phase 2 cutover.

Also adds HTTP-level test coverage for GET .../preview/calendar,
which had none before this -- it's what caught a stale photo_inlay
kwarg left over from the _build signature change that would have
TypeError'd on every request.
2026-07-24 09:38:10 -04:00

95 lines
4.2 KiB
Python

"""Maps named faces (from Immich's own face recognition/People feature)
onto their position in the final rendered frame, for the manage-button
overlay's named-face labels (see manage_overlay.py, which draws them).
No face detection or recognition happens here or anywhere else in this
project -- Immich's GET /api/faces?id={assetId} already returns each
detected face's bounding box plus a nullable `person` object (with a
`name`, if the user has identified them in Immich); this module only
does the coordinate math to place a label next to a *named* one.
"""
from __future__ import annotations
import io
from PIL import Image, ImageOps
from .image_pipeline import _has_bounding_box, _placement_transform, logical_render_size
# Not a memory constraint anymore (the overlay renders server-side now,
# not malloc'd per-label on the device) -- purely a legibility cap. A
# photo with a dozen named people would just be visual clutter regardless
# of what's rendering it.
MAX_LABELED_FACES = 6
def compute_face_labels(preview_bytes: bytes, faces: list[dict], display_mode: str,
orientation: str = "landscape", region: tuple[int, int, int, int] | None = None) -> list[dict]:
"""Returns up to MAX_LABELED_FACES [{"name", "x", "y"}], x/y in
logical (pre-rotation) frame space at each named face's bottom-center
point -- manage_overlay.compose() draws these directly onto the
logical-space image before it's rotated into native panel space, so
no rotation happens here (contrast with the old firmware-side
version, which drew post-rotation and needed logical_to_native).
Faces without an Immich-identified person name are skipped entirely.
preview_bytes must be the same preview image render_frame() used for
the currently-displayed frame, and display_mode/orientation must
match the settings that were active then -- otherwise the placement
computed here won't match what's actually on screen.
`region` is (x0, y0, w, h): where in the logical canvas the photo
actually landed, if not the whole thing -- e.g. a photo widget placed
in one corner of the panel rather than full-screen (see
routers/common.py's build_manage_content, which passes each photo
widget's own placement rect) -- without this a label would be placed
as if the photo filled the entire canvas, landing well off where the
widget actually is. None (the default) means the photo fills the
whole logical canvas.
The placement math matches render_frame()'s own composition step
exactly (see image_pipeline._placement_transform, shared so the two
can't drift apart).
"""
named = [face for face in faces if (face.get("person") or {}).get("name")]
if not named:
return []
if region is None:
logical_w, logical_h = logical_render_size(orientation)
region_x0, region_y0, target_w, target_h = 0, 0, logical_w, logical_h
else:
region_x0, region_y0, target_w, target_h = region
fitted = ImageOps.exif_transpose(Image.open(io.BytesIO(preview_bytes)).convert("RGB"))
scale_x, scale_y, offset_x, offset_y = _placement_transform(
fitted.width, fitted.height, target_w, target_h, display_mode, faces
)
labels = []
for face in named[:MAX_LABELED_FACES]:
if not _has_bounding_box(face):
continue
face_w = face.get("imageWidth") or fitted.width
face_h = face.get("imageHeight") or fitted.height
img_scale_x = fitted.width / face_w
img_scale_y = fitted.height / face_h
center_x = (face["boundingBoxX1"] + face["boundingBoxX2"]) / 2 * img_scale_x
bottom_y = face["boundingBoxY2"] * img_scale_y
# Relative to the region's own origin first (matches
# _placement_transform's target_w/target_h space), then shifted
# into full-canvas coordinates.
region_x = center_x * scale_x + offset_x
region_y = bottom_y * scale_y + offset_y
if not (0 <= region_x <= target_w and 0 <= region_y <= target_h):
continue # this face got cropped out of the region entirely
labels.append({"name": face["person"]["name"],
"x": int(region_x + region_x0), "y": int(region_y + region_y0)})
return labels