# Referring Expression Segmentation on PBench

Source: [Referring Expression Segmentation on PBench](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-31T08:40:30.728Z

Evaluation of Referring Expression Segmentation performance on the PBench dataset, reporting accuracy across fine-grained categories such as attribute understanding, spatial relations, and OCR guidance.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Referring Expression Segmentation |
| Dataset (including any stated split) | PBench |
| Metric | L0: Simple Objects Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 8 of 8 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [L0: Simple Objects Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=L0%3A%20Simple%20Objects%20Accuracy>): Higher is better; unit: PERCENT.

- [L1: Attribute Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=L1%3A%20Attribute%20Accuracy>): Higher is better; unit: PERCENT.

- [L2: OCR Guided Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=L2%3A%20OCR%20Guided%20Accuracy>): Higher is better; unit: PERCENT.

- [L3: Spatial Understanding Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=L3%3A%20Spatial%20Understanding%20Accuracy>): Higher is better; unit: PERCENT.

- [L4: Relation Binding Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=L4%3A%20Relation%20Binding%20Accuracy>): Higher is better; unit: PERCENT.

- [Dense Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=Dense%20Accuracy>): Higher is better; unit: PERCENT.

- [Average Accuracy](<https://www.sota2.com/research/sota/referring-expression-segmentation-on-pbench.md?metric=Average%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | L0: Simple Objects Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen3-VL-30B | 69.2 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Moondream2-2B | 68 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Qwen3-VL-4B | 67.3 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Qwen3-VL-8B | 65.6 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Falcon Perception | 65.1 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| SAM 3-0.9B | 64.3 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Moondream3-9B | 63.2 | Not recorded | 69cb54fe0636e5bf4afc9e7c |
| Qwen3-VL-2B | 54.1 | Not recorded | 69cb54fe0636e5bf4afc9e7c |

### Qwen3-VL-30B

L0: Simple Objects Accuracy: 69.2. Result ID: 69cb886b47f34ce924b3dcd1.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"30B"}. This record does not assert independent reproduction.

### Moondream2-2B

L0: Simple Objects Accuracy: 68. Result ID: 69cb886b47f34ce924b3dcd2.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

Recorded evaluation configuration: {"Parameters":"2B"}. This record does not assert independent reproduction.

### Qwen3-VL-4B

L0: Simple Objects Accuracy: 67.3. Result ID: 69cb886b47f34ce924b3dccf.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"4B"}. This record does not assert independent reproduction.

### Qwen3-VL-8B

L0: Simple Objects Accuracy: 65.6. Result ID: 69cb886b47f34ce924b3dcd0.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"8B"}. This record does not assert independent reproduction.

### Falcon Perception

L0: Simple Objects Accuracy: 65.1. Result ID: 69cb886b47f34ce924b3dcd5.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

Recorded evaluation configuration: {"Parameters":"0.6B"}. This record does not assert independent reproduction.

### SAM 3-0.9B

L0: Simple Objects Accuracy: 64.3. Result ID: 69cb886b47f34ce924b3dcd4.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6923e69bd6b0dd334292f0ad.md>)

[Implementation](<https://github.com/facebookresearch/sam3>)

Recorded evaluation configuration: {"Parameters":"0.9B"}. This record does not assert independent reproduction.

### Moondream3-9B

L0: Simple Objects Accuracy: 63.2. Result ID: 69cb886b47f34ce924b3dcd3.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

Recorded evaluation configuration: {"Parameters":"9B"}. This record does not assert independent reproduction.

### Qwen3-VL-2B

L0: Simple Objects Accuracy: 54.1. Result ID: 69cb886b47f34ce924b3dcce.

[Reporting paper](<https://www.sota2.com/research/paper/69cb54fe0636e5bf4afc9e7c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"2B"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.