File size: 2,697 Bytes
ff6dd8c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e44afa3
 
ff6dd8c
 
4185985
ff6dd8c
4185985
 
ff6dd8c
 
 
 
 
df2cec0
 
ff6dd8c
1dd8d94
e14c71e
 
 
1dd8d94
207e769
1dd8d94
 
 
fded5ec
c38ee1b
3e5d5d1
1dd8d94
 
 
 
 
f64fa8f
773311c
1dd8d94
 
 
4b4b16e
c521ddc
4b4b16e
1dd8d94
 
 
4b4b16e
1dd8d94
 
 
 
 
 
 
 
 
 
 
4b4b16e
4a49361
 
 
 
1dd8d94
 
 
 
4b4b16e
af37fc8
4b4b16e
f37bbb0
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
---
dataset_info:
  features:
  - name: video
    dtype: string
  - name: videoType
    dtype: string
  - name: question
    dtype: string
  - name: options
    sequence: string
  - name: correctAnswer
    dtype: string
  - name: abilityType_L2
    dtype: string
  - name: abilityType_L3
    dtype: string
  - name: question_idx
    dtype: int64
  splits:
  - name: test
    num_bytes: 1135911
    num_examples: 1257
  download_size: 586803
  dataset_size: 1135911
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
task_categories:
  - video-text-to-text
---

<p align="center">
    <img src="./figs/LOGO_v3.png" width="30%" height="30%">
</p>

# MMR-V: *Can MLLMs Think with Video?* A Benchmark for Multimodal Deep Reasoning in Videos


<p align="center">
  <a href="https://arxiv.org/abs/2506.04141"> 📝 Paper</a></a> |
  <a href="https://github.com/GaryStack/MMR-V"> 💻 Code</a> |
  <a href="https://mmr-v.github.io/"> 🏠 Homepage</a>
</p>




## 👀 MMR-V Data Card ("Think with Video")
The sequential structure of videos poses a challenge to the ability of multimodal large language models (MLLMs) to locate multi-frame evidence🕵️ and conduct multimodal reasoning. However, existing video benchmarks mainly focus on understanding tasks, which only require models to match frames mentioned in the question (referred to as "question frame") and perceive a few adjacent frames. To address this gap, we propose **MMR-V: A Benchmark for Multimodal Deep Reasoning in Videos**. MMR-V consists of **317** videos and **1,257** tasks. Models like o3 and o4-mini have achieved impressive results on image reasoning tasks by leveraging tool use to enable 🕵️evidence mining on images. Similarly, tasks in MMR-V require models to perform in-depth reasoning and analysis over visual information from different frames of a video, challenging their ability to 🕵️**think with video and mine evidence across long-range multi-frame**.

## 🎬 MMR-V Task Examples

<p align="center">
    <img src="./figs/data_example_intro_v4_5_16.png" width="80%" height="80%">
</p>

## 📚 Evaluation

1. Load the MMR-V Videos

```shell
huggingface-cli download JokerJan/MMR-VBench --repo-type dataset --local-dir MMR-V --local-dir-use-symlinks False
```
2. Extract videos from the `.tar` files:

```shell
cat videos.tar.part.* > videos.tar
tar -xvf videos.tar
```

3. Load MMR-V Benchmark:

```shell
samples = load_dataset("JokerJan/MMR-VBench", split='test')
```

## 🎯 Experiment Results


<p align="center">
    <img src="./figs/main.png" width="70%" height="70%">
</p>


## Dataset Details

Curated by: MMR-V Team

Language(s) (NLP): English

License: CC-BY 4.0