Referring Expression Segmentation
The task aims at labeling the pixels of an image or video that represent an object instance referred by a linguistic expression. In particular, the referring expression (RE) must allow the identification of an individual object in a discourse or scene (the referent). REs unambiguously identify the target instance.
Papers
Showing 1–10 of 145 papers
All datasetsRefCoCo valRefCOCO testARefer-YouTube-VOS (2021 public validation)RefCOCO+ test BA2D SentencesRefCOCOg-valJ-HMDBDAVIS 2017 (val)RefCOCOg-testRefCOCO testBPhraseCutRefCOCO
Benchmark Results
| # | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| 1 | UNINEXT-H | J&F 1st frame | 72.5 | — | Unverified |
| 2 | HyperSeg | J&F 1st frame | 71.2 | — | Unverified |
| 3 | DEVA (ReferFormer) | J&F 1st frame | 66.3 | — | Unverified |
| 4 | HTR | J&F 1st frame | 65.6 | — | Unverified |
| 5 | VATEX | J&F score | 65.4 | — | Unverified |
| 6 | SgMg | J&F 1st frame | 63.3 | — | Unverified |
| 7 | SafaRi-B | J&F 1st frame | 61.3 | — | Unverified |
| 8 | ReferFormer | J&F 1st frame | 61.1 | — | Unverified |
| 9 | PolyFormer-B | J&F 1st frame | 60.9 | — | Unverified |
| 10 | UniVS(Swin-L) | J&F Full video | 59.4 | — | Unverified |