Skip to content
← Models
Open weightsUIUC / Adobe Research/10 trustMetrics-checked

Cutie

Object-level memory reading for video segmentation: remembers the object, not just pixels.

Visit official page →

About

Cutie is the follow-up to XMem from the same UIUC group with Adobe Research. Where XMem does bottom-up pixel-level memory matching, which gets noisy around distractors, Cutie adds top-down object-level memory reading: a small set of object queries summarizes the target and iteratively interacts with pixel features through an object transformer, with foreground-background masked attention keeping the semantics clean.

A CVPR 2024 highlight, the authors report Cutie-base at 68.3 J&F on MOSE (8.7 over XMem at similar speed, 4.2 over DeAOT at 3x the speed), 88.8 J&F on DAVIS-2017 val, and 86.5 G on YouTubeVOS-2019 val. The license was not explicitly confirmed in the sources checked, so verify before commercial use; the bundled GUI includes RITM code under its own license.

Specs

Context window

n/a

Pricing

Free (open weights; license unconfirmed)

Free quota

Free — self-hosted weights; official checkpoints on GitHub

API access

No official API; GitHub checkpoints; interactive GUI demo

Weights

Open weights

Released

Not disclosed

Benchmarks

Author-reported (CVPR 2024 Highlight): Cutie-base 68.3 J&F MOSE (+8.7 over XMem), 88.8 J&F DAVIS-2017 val, 86.5 G YouTubeVOS-2019 val

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

video-object-segmentationvideo-editingrotoscopingmulti-object-video

Trust score — 6/10

/10 trustMetrics-checked

Scored 6/10 (metrics tier, metrics-v1). Evidence: CVPR 2024 highlight paper (arXiv:2310.12982) with MOSE/DAVIS/YouTubeVOS numbers confirmed via the paper table; official repo confirmed at github.com/hkchengrex/Cutie. Strengths: best-in-class VOS accuracy at release, faster than DeAOT, GUI demo included. Docked for: (1) license not explicitly confirmed in the sources checked; (2) bundled RITM GUI code under a separate license adds compliance surface. License: not explicitly confirmed — flag for review.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
6.0/10
Openness · license + docs
8.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · CVPR 2024 highlight (arXiv 2310.12982, Oct 2023)
  • · Cutie-base 68.3 J&F MOSE (+8.7 over XMem), 88.8 J&F DAVIS-2017 val, author-reported
  • · Object-level memory reading via query-based object transformer

Sources