{{ templates|length }} vases currently in the catalogue
{% for t in templates %}
{{ t.name }}
{% endfor %}
Find your match
Upload a photo — background removal and all four matchers run automatically.
Removing background…
Results
Weighted final match
Flower check
Comparing flowers…
Flower count
Flower Vase similarity (CLIP)
Flower count
Optional and separate from matching — two
independent methods, shown side by side. SAM (Segment Anything) finds individual
flower-shaped regions and color-clusters them as a rough guide to how many
distinct kinds are in the bouquet; YOLO-World is prompted to recognize "flower",
"vase", and "ribbon"/"bow" by name, which is also used to stop the vase and any
ribbon from being miscounted as flowers by SAM. Not run automatically: it
temporarily unloads the matching models to free GPU memory, so it can take a
little longer the first time.
Loading SAM + YOLO-World and segmenting your photo…
SAM — instance segmentation
YOLO-World — open-vocabulary detection
Vase comparison — DINOv2 + CLIP
Your upload
→
DINOv2 similarity
CLIP similarity
Combined
Color space
How closely each template's color palette matches your upload's —
separate from the SIFT/ORB/SuperGlue/LoFTR match scores above, though it does contribute
its own share to the weighted final match below.
Shape matching
How closely each template's overall silhouette matches your upload's —
independent of color and texture, and it doesn't factor into the final verdict either.
Uses Hu-moment contour matching plus a direct silhouette-overlap check.
Texture matching
How closely each template's surface texture matches your upload's —
independent of color and shape, and it doesn't factor into the final verdict either.
Uses Local Binary Patterns plus GLCM (Haralick) statistical texture features.
Color family grid
Your upload vs ,
each broken into its own dynamically-discovered color regions, matched up region by region.
AI verification
Cross-checked against by an
external vision model, comparing flowers, vase, and ribbon detail by detail.
 }})