Vesti
Can you replace one piece without losing the outfit?
An outfit-search project exploring garment segmentation, visual similarity and alternative pieces.
The problem
An outfit reference gives useful context, but a person may only want to replace one garment. Comparing whole images can lose that intent.
The approach
- The planned flow starts with a reference image and a selected piece.
- Use SAM 3.1 for garment segmentation and Marqo FashionSigLIP for visual comparison.
- Build the candidate-matching stage with Jev to connect the selected garment to available alternatives.
What you can inspect
The proposed pipeline is documented below. There is no public demo or measured result to present yet.
Scope & limitations
Work in progress. The intended flow is a design direction, not a claim that the full experience is shipped.
Technical pipeline
The pipeline we’re building. Open a stage to go deeper.
Outfit inputReference image · piece selection
The planned experience starts with an outfit image. A person chooses the piece they want to replace while keeping the rest of the outfit as context.
Garment segmentationSAM 3.1
The first model stage separates the clothing items in the image. Isolating a garment gives the comparison stage a focused reference instead of asking it to compare the entire outfit.
Visual comparisonMarqo FashionSigLIP
Marqo FashionSigLIP compares clothing visually to find similar pieces. This stage identifies resemblance to the selected garment; it is separate from deciding which available alternative works for the look.
Candidate matchingBuilding with Jev
We’re building the matching stage with Jev to connect the reference piece to available options. The intended result is a set of replacements someone can explore without restarting the outfit search.