# Reasoning-driven segmentation on SegEarth R2 (train) Source: [Reasoning-driven segmentation on SegEarth R2 (train)]() Collection index: [llms.txt]() Data updated: 2026-03-05T09:11:09.456Z Evaluation of Reasoning-driven segmentation performance on the training split of the SegEarth R2 dataset, focusing on model-specific reasoning metrics like Faithfulness, Localization, and Robustness for Qwen models and user evaluations. 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 | Reasoning-driven segmentation | | Dataset (including any stated split) | SegEarth R2 (train) | | Metric | Faithfulness (Qwen-8B) | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 14 of 14 | | 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 - [Faithfulness (Qwen-8B)](): Higher is better; unit: SCALAR. - [Localization (Qwen-8B)](): Higher is better; unit: SCALAR. - [Robustness (Qwen-8B)](): Higher is better; unit: SCALAR. - [Faithfulness (Qwen-32B)](): Higher is better; unit: SCALAR. - [Localization (Qwen-32B)](): Higher is better; unit: SCALAR. - [Robustness (Qwen-32B)](): Higher is better; unit: SCALAR. - [Faithfulness (User)](): Higher is better; unit: SCALAR. - [Localization (User)](): Higher is better; unit: SCALAR. - [Robustness (User)](): Higher is better; unit: SCALAR. ## Recorded results | Method | Faithfulness (Qwen-8B) | Result date | Report paper | | --- | --- | --- | --- | | LISA-7B | 2.12 | Not recorded | 69a90e162d603e6de3e408ee | | GeoSeg | 1.78 | Not recorded | 69a90e162d603e6de3e408ee | | CLIP Surgery | 1.57 | Not recorded | 69a90e162d603e6de3e408ee | | Grounded SAM | 1.42 | Not recorded | 69a90e162d603e6de3e408ee | | SAM3 | 1.39 | Not recorded | 69a90e162d603e6de3e408ee | | BAGEL | 1.26 | Not recorded | 69a90e162d603e6de3e408ee | | GSVA | 1.25 | Not recorded | 69a90e162d603e6de3e408ee | | LISA++ | 1.16 | Not recorded | 69a90e162d603e6de3e408ee | | UniWorld | 1.15 | Not recorded | 69a90e162d603e6de3e408ee | | OmniGen2 | 1.14 | Not recorded | 69a90e162d603e6de3e408ee | | Qwen-Edit | 1.12 | Not recorded | 69a90e162d603e6de3e408ee | | PixelLM | 1.1 | Not recorded | 69a90e162d603e6de3e408ee | | Hunyuan | 1.04 | Not recorded | 69a90e162d603e6de3e408ee | | GLAMM | 1 | Not recorded | 69a90e162d603e6de3e408ee | ### LISA-7B Faithfulness (Qwen-8B): 2.12. Result ID: 69a948966dbb7075b952f7ee. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Reasoning segmentation"}. This record does not assert independent reproduction. ### GeoSeg Faithfulness (Qwen-8B): 1.78. Result ID: 69a948966dbb7075b952f7f6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Ours"}. This record does not assert independent reproduction. ### CLIP Surgery Faithfulness (Qwen-8B): 1.57. Result ID: 69a948966dbb7075b952f7eb. [Reporting paper]() Recorded evaluation configuration: {"Category":"Generalist segmentation"}. This record does not assert independent reproduction. ### Grounded SAM Faithfulness (Qwen-8B): 1.42. Result ID: 69a948966dbb7075b952f7ea. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Generalist segmentation"}. This record does not assert independent reproduction. ### SAM3 Faithfulness (Qwen-8B): 1.39. Result ID: 69a948966dbb7075b952f7e9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Generalist segmentation"}. This record does not assert independent reproduction. ### BAGEL Faithfulness (Qwen-8B): 1.26. Result ID: 69a948966dbb7075b952f7f3. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Open-source MLLMs"}. This record does not assert independent reproduction. ### GSVA Faithfulness (Qwen-8B): 1.25. Result ID: 69a948966dbb7075b952f7ed. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Reasoning segmentation"}. This record does not assert independent reproduction. ### LISA++ Faithfulness (Qwen-8B): 1.16. Result ID: 69a948966dbb7075b952f7ef. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Reasoning segmentation"}. This record does not assert independent reproduction. ### UniWorld Faithfulness (Qwen-8B): 1.15. Result ID: 69a948966dbb7075b952f7f4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Open-source MLLMs"}. This record does not assert independent reproduction. ### OmniGen2 Faithfulness (Qwen-8B): 1.14. Result ID: 69a948966dbb7075b952f7f2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Open-source MLLMs"}. This record does not assert independent reproduction. ### Qwen-Edit Faithfulness (Qwen-8B): 1.12. Result ID: 69a948966dbb7075b952f7f1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Open-source MLLMs"}. This record does not assert independent reproduction. ### PixelLM Faithfulness (Qwen-8B): 1.1. Result ID: 69a948966dbb7075b952f7f0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Reasoning segmentation"}. This record does not assert independent reproduction. ### Hunyuan Faithfulness (Qwen-8B): 1.04. Result ID: 69a948966dbb7075b952f7f5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Category":"Open-source MLLMs"}. This record does not assert independent reproduction. ### GLAMM Faithfulness (Qwen-8B): 1. Result ID: 69a948966dbb7075b952f7ec. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Category":"Reasoning segmentation"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.