# Fine-grained Video Spatial Reasoning on VSI-Bench

Source: [Fine-grained Video Spatial Reasoning on VSI-Bench](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench>)

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

Data updated: 2026-02-26T05:36:23.374Z

Evaluation of fine-grained spatial reasoning capabilities using video inputs on the VSI-Bench dataset, reporting detailed component metrics such as object counts, relative distances, and route planning accuracy.

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 | Fine-grained video-based spatial reasoning |
| Dataset (including any stated split) | VSI-Bench |
| Metric | Avg Score |
| Unit | PERCENT |
| 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

- [Avg Score](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Avg%20Score>): Higher is better; unit: PERCENT.

- [Obj Count Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Obj%20Count%20Acc>): Higher is better; unit: PERCENT.

- [Abs Dist Err](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Abs%20Dist%20Err>): Lower is better; unit: SCALAR.

- [Obj Size Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Obj%20Size%20Acc>): Higher is better; unit: PERCENT.

- [Room Size Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Room%20Size%20Acc>): Higher is better; unit: PERCENT.

- [Rel Dist Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Rel%20Dist%20Acc>): Higher is better; unit: PERCENT.

- [Rel Dir Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Rel%20Dir%20Acc>): Higher is better; unit: PERCENT.

- [Route Plan Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Route%20Plan%20Acc>): Higher is better; unit: PERCENT.

- [Appr Ord Acc](<https://www.sota2.com/research/sota/fine-grained-video-spatial-reasoning-on-vsi-bench.md?metric=Appr%20Ord%20Acc>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Avg Score | Result date | Report paper |
| --- | --- | --- | --- |
| VST-7B-SFT | 60.6 | Not recorded | 6965d1a46343fa78c832381c |
| SmoothOp-7B | 60 | Not recorded | 6965d1a46343fa78c832381c |
| VST-3B-SFT | 57.9 | Not recorded | 6965d1a46343fa78c832381c |
| SmoothOp-3B | 53.9 | Not recorded | 6965d1a46343fa78c832381c |
| InternVL-Spatial-8B | 52.3 | Not recorded | 6965d1a46343fa78c832381c |
| SpatialLadder-3B | 45.7 | Not recorded | 6965d1a46343fa78c832381c |
| Gemini-1.5 Pro | 45.4 | Not recorded | 6965d1a46343fa78c832381c |
| ViLaSR-7B | 45.4 | Not recorded | 6965d1a46343fa78c832381c |
| SpaceR-7B | 43.5 | Not recorded | 6965d1a46343fa78c832381c |
| LLaVA-Video-7B | 35.6 | Not recorded | 6965d1a46343fa78c832381c |
| GPT-4o | 34 | Not recorded | 6965d1a46343fa78c832381c |
| Qwen2.5-VL-7B | 32.7 | Not recorded | 6965d1a46343fa78c832381c |
| LLaVA-OneVision-7B | 32.4 | Not recorded | 6965d1a46343fa78c832381c |

### VST-7B-SFT

Avg Score: 60.6. Result ID: 6997c8ba5698665ec41fb871.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691171712828b7cc6937d56e.md>)

[Implementation](<https://github.com/Yangr116/VST>)

Recorded evaluation configuration: {"Data Scale":"4.1M","Training Phases":"1 (S)"}. This record does not assert independent reproduction.

### SmoothOp-7B

Avg Score: 60. Result ID: 6997c8ba5698665ec41fb873.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

Recorded evaluation configuration: {"Data Scale":"50K","Training Phases":"1 (R)"}. This record does not assert independent reproduction.

### VST-3B-SFT

Avg Score: 57.9. Result ID: 6997c8ba5698665ec41fb870.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/691171712828b7cc6937d56e.md>)

[Implementation](<https://github.com/Yangr116/VST>)

Recorded evaluation configuration: {"Data Scale":"4.1M","Training Phases":"1 (S)"}. This record does not assert independent reproduction.

### SmoothOp-3B

Avg Score: 53.9. Result ID: 6997c8ba5698665ec41fb872.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

Recorded evaluation configuration: {"Data Scale":"50K","Training Phases":"1 (R)"}. This record does not assert independent reproduction.

### InternVL-Spatial-8B

Avg Score: 52.3. Result ID: 6997c8ba5698665ec41fb86d.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71de1e209131cea5370d.md>)

[Implementation](<https://github.com/OpenGVLab/InternVL>)

Recorded evaluation configuration: {"Data Scale":"291K","Training Phases":"1 (S)"}. This record does not assert independent reproduction.

### SpatialLadder-3B

Avg Score: 45.7. Result ID: 6997c8ba5698665ec41fb86c.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

Recorded evaluation configuration: {"Data Scale":"26K","Training Phases":"3 (S+R)"}. This record does not assert independent reproduction.

### Gemini-1.5 Pro

Avg Score: 45.4. Result ID: 6997c8ba5698665ec41fb867.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76361e209131cea6be11.md>)

Recorded evaluation configuration: {"Data Scale":"-","Training Phases":"-"}. This record does not assert independent reproduction.

### ViLaSR-7B

Avg Score: 45.4. Result ID: 6997c8ba5698665ec41fb86f.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79831e209131cea7defc.md>)

Recorded evaluation configuration: {"Data Scale":"81K","Training Phases":"2 (S+R)"}. This record does not assert independent reproduction.

### SpaceR-7B

Avg Score: 43.5. Result ID: 6997c8ba5698665ec41fb86e.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a41e209131cea798a1.md>)

Recorded evaluation configuration: {"Data Scale":"151K","Training Phases":"2 (S+R)"}. This record does not assert independent reproduction.

### LLaVA-Video-7B

Avg Score: 35.6. Result ID: 6997c8ba5698665ec41fb869.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

Recorded evaluation configuration: {"Data Scale":"-","Training Phases":"-"}. This record does not assert independent reproduction.

### GPT-4o

Avg Score: 34. Result ID: 6997c8ba5698665ec41fb866.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75681e209131cea67a98.md>)

Recorded evaluation configuration: {"Data Scale":"-","Training Phases":"-"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B

Avg Score: 32.7. Result ID: 6997c8ba5698665ec41fb86a.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75151e209131cea65976.md>)

[Implementation](<https://github.com/QwenLM/Qwen2-VL>)

Recorded evaluation configuration: {"Data Scale":"-","Training Phases":"-"}. This record does not assert independent reproduction.

### LLaVA-OneVision-7B

Avg Score: 32.4. Result ID: 6997c8ba5698665ec41fb868.

[Reporting paper](<https://www.sota2.com/research/paper/6965d1a46343fa78c832381c.md>)

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

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

Recorded evaluation configuration: {"Data Scale":"-","Training Phases":"-"}. This record does not assert independent reproduction.

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