# Perspective-aware spatial reasoning on COMFORT Visual Illusions

Source: [Perspective-aware spatial reasoning on COMFORT Visual Illusions](<https://www.sota2.com/research/sota/perspective-aware-spatial-reasoning-on-comfort-visual-illusions>)

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

Data updated: 2026-02-26T05:41:09.107Z

Evaluation of model performance on the COMFORT Visual Illusions dataset, focusing on perspective-aware spatial reasoning capabilities, measured by directional accuracy (left/right, front/behind) and depth perception (closer).

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 | Perspective-aware spatial reasoning |
| Dataset (including any stated split) | COMFORT Visual Illusions |
| Metric | Directional Accuracy (Left/Right) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 19 of 19 |
| 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

- [Directional Accuracy (Left/Right)](<https://www.sota2.com/research/sota/perspective-aware-spatial-reasoning-on-comfort-visual-illusions.md?metric=Directional%20Accuracy%20(Left%2FRight)>): Higher is better; unit: PERCENT.

- [Directional Accuracy (Front/Behind)](<https://www.sota2.com/research/sota/perspective-aware-spatial-reasoning-on-comfort-visual-illusions.md?metric=Directional%20Accuracy%20(Front%2FBehind)>): Higher is better; unit: PERCENT.

- [Depth Perception Accuracy (Closer)](<https://www.sota2.com/research/sota/perspective-aware-spatial-reasoning-on-comfort-visual-illusions.md?metric=Depth%20Perception%20Accuracy%20(Closer)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Directional Accuracy (Left/Right) | Result date | Report paper |
| --- | --- | --- | --- |
| SymPL | 95.38 | Not recorded | 699d30776017c4460ffab35c |
| APC-Num | 84.31 | Not recorded | 699d30776017c4460ffab35c |
| APC-Vis | 76.75 | Not recorded | 699d30776017c4460ffab35c |
| LLaVa-OneVision | 72 | Not recorded | 699d30776017c4460ffab35c |
| LLaVa-NeXT | 58.19 | Not recorded | 699d30776017c4460ffab35c |
| Qwen2.5-VL | 58.06 | Not recorded | 699d30776017c4460ffab35c |
| Qwen2.5-VL | 57 | Not recorded | 699d30776017c4460ffab35c |
| Gemini-2.5-Flash | 54.63 | Not recorded | 699d30776017c4460ffab35c |
| SpatialBot | 52.25 | Not recorded | 699d30776017c4460ffab35c |
| SD-VLM | 50 | Not recorded | 699d30776017c4460ffab35c |
| Random | 49.69 | Not recorded | 699d30776017c4460ffab35c |
| Qwen2.5-VL | 46.56 | Not recorded | 699d30776017c4460ffab35c |
| Molmo | 44.56 | Not recorded | 699d30776017c4460ffab35c |
| SpatialVLM | 44 | Not recorded | 699d30776017c4460ffab35c |
| GPT-5 | 43.63 | Not recorded | 699d30776017c4460ffab35c |
| SpatialRGPT | 42.31 | Not recorded | 699d30776017c4460ffab35c |
| Qwen2.5-VL | 40.94 | Not recorded | 699d30776017c4460ffab35c |
| Cambrian-1 | 35.06 | Not recorded | 699d30776017c4460ffab35c |
| SAT | 29.44 | Not recorded | 699d30776017c4460ffab35c |

### SymPL

Directional Accuracy (Left/Right): 95.38. Result ID: 699e38075698665ec4f470ff.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### APC-Num

Directional Accuracy (Left/Right): 84.31. Result ID: 699e38075698665ec4f470fd.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78311e209131cea77193.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### APC-Vis

Directional Accuracy (Left/Right): 76.75. Result ID: 699e38075698665ec4f470fe.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78311e209131cea77193.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### LLaVa-OneVision

Directional Accuracy (Left/Right): 72. Result ID: 699e38075698665ec4f470ef.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa756f1e209131cea67e58.md>)

[Implementation](<https://llava-vl.github.io/blog/llava-onevision>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### LLaVa-NeXT

Directional Accuracy (Left/Right): 58.19. Result ID: 699e38075698665ec4f470ee.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Qwen2.5-VL

Directional Accuracy (Left/Right): 58.06. Result ID: 699e38075698665ec4f470f7.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Prompting":"SCAFFOLD"}. This record does not assert independent reproduction.

### Qwen2.5-VL

Directional Accuracy (Left/Right): 57. Result ID: 699e38075698665ec4f470f1.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Gemini-2.5-Flash

Directional Accuracy (Left/Right): 54.63. Result ID: 699e38075698665ec4f470f4.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c311e209131cea8cef5.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SpatialBot

Directional Accuracy (Left/Right): 52.25. Result ID: 699e38075698665ec4f470fa.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77821e209131cea73b63.md>)

[Implementation](<https://github.com/BAAI-DCAI/SpatialBot>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SD-VLM

Directional Accuracy (Left/Right): 50. Result ID: 699e38075698665ec4f470fb.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b881e209131cea88f0d.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Random

Directional Accuracy (Left/Right): 49.69. Result ID: 699e38075698665ec4f470ed.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Qwen2.5-VL

Directional Accuracy (Left/Right): 46.56. Result ID: 699e38075698665ec4f470f6.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Prompting":"SoM"}. This record does not assert independent reproduction.

### Molmo

Directional Accuracy (Left/Right): 44.56. Result ID: 699e38075698665ec4f470f0.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76101e209131cea6b37c.md>)

[Implementation](<https://molmo.allenai.org/>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SpatialVLM

Directional Accuracy (Left/Right): 44. Result ID: 699e38075698665ec4f470f8.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71f91e209131cea53d86.md>)

[Implementation](<https://spatial-vlm.github.io/>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GPT-5

Directional Accuracy (Left/Right): 43.63. Result ID: 699e38075698665ec4f470f3.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SpatialRGPT

Directional Accuracy (Left/Right): 42.31. Result ID: 699e38075698665ec4f470f9.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75321e209131cea66538.md>)

[Implementation](<https://www.anjiecheng.me/SpatialRGPT>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Qwen2.5-VL

Directional Accuracy (Left/Right): 40.94. Result ID: 699e38075698665ec4f470f5.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Prompting":"CoT"}. This record does not assert independent reproduction.

### Cambrian-1

Directional Accuracy (Left/Right): 35.06. Result ID: 699e38075698665ec4f470f2.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75fa1e209131cea6ac2e.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SAT

Directional Accuracy (Left/Right): 29.44. Result ID: 699e38075698665ec4f470fc.

[Reporting paper](<https://www.sota2.com/research/paper/699d30776017c4460ffab35c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c91e209131cea75541.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

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