SEEM
Promptable segmentation that takes text, points, boxes, scribbles, or masks, alone or composed together.
About
SEEM (Segment Everything Everywhere All at Once) is a universal segmentation interface from UW-Madison and Microsoft Research. Instead of one prompt type, it accepts text, points, boxes, scribbles, masks, and even referring images, and learns a joint visual-semantic space so you can compose prompts (say, a box plus a text refinement) in a single forward pass. Learnable memory prompts retain segmentation history across interactive clicks.
One SEEM model covers interactive, generic, referring, and video object segmentation across nine datasets, and the authors report interactive results better than SAM-Base despite training on roughly 1/100th the segmentation data. Focal-T checkpoints score 50.6 PQ, 39.4 mAP, and 60.9 mIoU on COCO. Note: the repo license was not explicitly confirmed in the sources checked, so verify before commercial use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported: competitive across 9 datasets; beats SAM-Base on interactive segmentation with ~1/100 the data; Focal-T 50.6 COCO PQ / 39.4 mAP / 60.9 mIoU
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: NeurIPS 2023 paper (arXiv:2304.06718) with 9-dataset results confirmed via the official repo README checkpoint table. Strengths: most flexible prompting interface in segmentation, strong results with little training data. Docked for: (1) repo license not explicitly confirmed in the sources checked, commercial use unverified; (2) interactive GUI setup is fiddly and docs assume Linux. License: not explicitly confirmed — flag for review.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · NeurIPS 2023 (arXiv 2304.06718, Apr 2023)
- · Beats SAM-Base on interactive segmentation with ~1/100 the training data, author-reported
- · Composes text, points, boxes, scribbles, masks, and referring-image prompts