# Multi-Image Fine-Grained Question Answering on MIFG-QA

Source: [Multi-Image Fine-Grained Question Answering on MIFG-QA](<https://www.sota2.com/research/sota/multi-image-fine-grained-question-answering-on-mifg-qa>)

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

Data updated: 2026-02-26T05:34:21.095Z

Evaluation of model performance on the MIFG-QA dataset for Multi-Image Fine-Grained Question Answering tasks.

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 | Multi-Image Fine-Grained Question Answering |
| Dataset (including any stated split) | MIFG-QA |
| Metric | Accuracy (Nature) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [Accuracy (Nature)](<https://www.sota2.com/research/sota/multi-image-fine-grained-question-answering-on-mifg-qa.md?metric=Accuracy%20(Nature)>): Higher is better; unit: PERCENT.

- [Accuracy (Poster)](<https://www.sota2.com/research/sota/multi-image-fine-grained-question-answering-on-mifg-qa.md?metric=Accuracy%20(Poster)>): Higher is better; unit: PERCENT.

- [Average Accuracy](<https://www.sota2.com/research/sota/multi-image-fine-grained-question-answering-on-mifg-qa.md?metric=Average%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (Nature) | Result date | Report paper |
| --- | --- | --- | --- |
| IMAgent | 50.24 | Not recorded | 693a50176cfd488761a54aa5 |
| GPT-4o | 49.26 | Not recorded | 693a50176cfd488761a54aa5 |
| Pixel Reasoner | 41.87 | Not recorded | 693a50176cfd488761a54aa5 |
| Deepeyes | 38.75 | Not recorded | 693a50176cfd488761a54aa5 |
| LLaVA-OneVision | 37.93 | Not recorded | 693a50176cfd488761a54aa5 |
| Qwen2.5-VL | 36.94 | Not recorded | 693a50176cfd488761a54aa5 |
| Qwen2.5-VL-CoT | 36.35 | Not recorded | 693a50176cfd488761a54aa5 |

### IMAgent

Accuracy (Nature): 50.24. Result ID: 699889005698665ec4ea800f.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

Recorded evaluation configuration: {"E2E":true,"Param":"7B"}. This record does not assert independent reproduction.

### GPT-4o

Accuracy (Nature): 49.26. Result ID: 699889005698665ec4ea8009.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

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

Recorded evaluation configuration: {"E2E":null,"Param":null}. This record does not assert independent reproduction.

### Pixel Reasoner

Accuracy (Nature): 41.87. Result ID: 699889005698665ec4ea800e.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c71e209131cea7a774.md>)

Recorded evaluation configuration: {"E2E":false,"Param":"7B"}. This record does not assert independent reproduction.

### Deepeyes

Accuracy (Nature): 38.75. Result ID: 699889005698665ec4ea800d.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c71e209131cea7a75e.md>)

Recorded evaluation configuration: {"E2E":true,"Param":"7B"}. This record does not assert independent reproduction.

### LLaVA-OneVision

Accuracy (Nature): 37.93. Result ID: 699889005698665ec4ea800a.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

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

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

Recorded evaluation configuration: {"E2E":null,"Param":"7B"}. This record does not assert independent reproduction.

### Qwen2.5-VL

Accuracy (Nature): 36.94. Result ID: 699889005698665ec4ea800b.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

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

Recorded evaluation configuration: {"E2E":null,"Param":"7B"}. This record does not assert independent reproduction.

### Qwen2.5-VL-CoT

Accuracy (Nature): 36.35. Result ID: 699889005698665ec4ea800c.

[Reporting paper](<https://www.sota2.com/research/paper/693a50176cfd488761a54aa5.md>)

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

Recorded evaluation configuration: {"E2E":null,"Param":"7B"}. This record does not assert independent reproduction.

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