Dataset Preview
Full Screen
The full dataset viewer is not available (click to read why). Only showing a preview of the rows.
The dataset generation failed because of a cast error
Error code:   DatasetGenerationCastError
Exception:    DatasetGenerationCastError
Message:      An error occurred while generating the dataset

All the data files must have the same columns, but at some point there are 4 new columns ({'dimensions', 'video_name', 'video_type', 'question'}) and 2 missing columns ({'org_index', 'answer'}).

This happened while the json dataset builder was generating data using

hf://datasets/opencompass/MMBench-Video/MMBench-Video_q.json (at revision 53b6cef24b62aadda40b4144e6c11ea824882d6c)

Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
                  writer.write_table(table)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/arrow_writer.py", line 580, in write_table
                  pa_table = table_cast(pa_table, self._schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/table.py", line 2292, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/table.py", line 2240, in cast_table_to_schema
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              video_name: string
              question_id: string
              question: string
              dimensions: string
              video_type: string
              -- schema metadata --
              pandas: '{"index_columns": [], "column_indexes": [], "columns": [{"name":' + 718
              to
              {'answer': Value(dtype='string', id=None), 'org_index': Value(dtype='int64', id=None), 'question_id': Value(dtype='string', id=None)}
              because column names don't match
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1387, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 980, in stream_convert_to_parquet
                  builder._prepare_split(
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1740, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/builder.py", line 1871, in _prepare_split_single
                  raise DatasetGenerationCastError.from_cast_error(
              datasets.exceptions.DatasetGenerationCastError: An error occurred while generating the dataset
              
              All the data files must have the same columns, but at some point there are 4 new columns ({'dimensions', 'video_name', 'video_type', 'question'}) and 2 missing columns ({'org_index', 'answer'}).
              
              This happened while the json dataset builder was generating data using
              
              hf://datasets/opencompass/MMBench-Video/MMBench-Video_q.json (at revision 53b6cef24b62aadda40b4144e6c11ea824882d6c)
              
              Please either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

answer
string
org_index
int64
question_id
string
Palmer.
0
wZxzBvAgqxc_0
3
1
wZxzBvAgqxc_1
Yes. The second goal in the video was originally shot by a Red team player, but the Blue team defenseman knocked the ball into his own goal.
2
wZxzBvAgqxc_2
This video is a key moments replay of a football game. The team in red jerseys scored two goals, and the team in blue jerseys scored one goal.
3
wZxzBvAgqxc_3
Paypal honey
4
lKNB3ZeTYiI_processed_0
sound-proofing earmuffs
5
lKNB3ZeTYiI_processed_1
It helps you save money when shopping online!
6
lKNB3ZeTYiI_processed_2
He was improving some item recipes in the game Minecraft.
7
4jOk3ajqJ2s_processed_0
Because the output of the original recipe is really annoying.
8
4jOk3ajqJ2s_processed_1
In the improved recipe, each stone block yields 72 stone buttons, therefore, 5 stone blocks yield 360. (5 x 72 = 360)
9
4jOk3ajqJ2s_processed_2
There is no way to know based on the video content.
10
4jOk3ajqJ2s_processed_3
7
11
rtfS6PQEot0_0
It's called "We Are More".
12
rtfS6PQEot0_1
Because other champions can block the impact of her charge.
13
rtfS6PQEot0_2
Shurima
14
rtfS6PQEot0_3
The name of the person that started the vote is no longer revealed.
15
qsqGAaEV3lw_0
He/she will not be able to start a vote, since a player can't initiate 2 votes within 6 minutes.
16
qsqGAaEV3lw_1
It will automatically close.
17
qsqGAaEV3lw_2
I don't know, since there are only 4 points mentioned in the video.
18
qsqGAaEV3lw_3
It's called "Royal Maelstrom".
19
lZC0QV4N3OU_processed_0
4
20
lZC0QV4N3OU_processed_1
Once has filled her stomach and assumed her ture form (after using R).
21
lZC0QV4N3OU_processed_2
I don't know, since there are only 5 ability mentioned in the video.
22
lZC0QV4N3OU_processed_3
Tranquilizer Rifle and Steel net gun.
23
6MSJsDyFxcQ_0
Sushi and Seafood Cuisine
24
6MSJsDyFxcQ_1
Seahorses
25
6MSJsDyFxcQ_2
It was on fire.
26
5zCc934KAi0_0
He jumped into the straw.
27
5zCc934KAi0_1
An assassin or a warrior.
28
5zCc934KAi0_2
He used a bow and arrow.
29
hSZLHJRZUOg_processed_0
There is no way to know based on the video content.
30
hSZLHJRZUOg_processed_1
19.3 tonnes
31
ZRrYCyEc9_M_0
A single airplane.
32
ZRrYCyEc9_M_1
Because you can milk an infinite amount of milk out of a cow, and the cow needs nothing to produce it.
33
ZRrYCyEc9_M_2
Cleaning Master and Drinking Serving Master
34
vTxcVxD3Uk0_0
17101
35
vTxcVxD3Uk0_1
825
36
vTxcVxD3Uk0_2
Winter.
37
aR-KAldshAE_0
A pair of metal gloves.
38
aR-KAldshAE_1
League of Legends
39
aR-KAldshAE_2
Poseidon's Trident
40
vNFSIHdsMT8_processed_0
A actual working plane.
41
Ak1eEUNrpgo_processed_0
Four.
42
Ak1eEUNrpgo_processed_1
There is no way to know based on the video content.
43
Ak1eEUNrpgo_processed_2
Green.
44
Uq9jqhSOUQQ_0
Bandito.
45
Uq9jqhSOUQQ_1
There is no way to know based on the video content.
46
Uq9jqhSOUQQ_2
Pink or purple
47
hIyrb8PeGXs_0
Poison.
48
hIyrb8PeGXs_1
There is no way to know based on the video content.
49
hIyrb8PeGXs_2
It can break through floors and wooden doors.
50
Qnyb73rf7gM_processed_0
4
51
Qnyb73rf7gM_processed_1
Red.
52
Qnyb73rf7gM_processed_2
There is no way to know based on the video content.
53
Qnyb73rf7gM_processed_3
A tower.
54
ka5t4Msfras_0
A helicopter
55
ka5t4Msfras_1
Yes, it's a woman.
56
ka5t4Msfras_2
Kitchen.
57
LqNCz7q0bf4_processed_0
Sunrise Bar.
58
LqNCz7q0bf4_processed_1
League of Legends and Minecraft.
59
dSHcCllTCzY_0
It's a good bet that it is.
60
dSHcCllTCzY_1
125
61
9mKJtnY3qgQ_processed_0
Red.
62
9mKJtnY3qgQ_processed_1
07:42.7
63
X-miELlaEYw_0
Realme
64
UBV5mJuJ1xg_0
Liu Chuanzhi
65
UBV5mJuJ1xg_1
Wiko
66
UBV5mJuJ1xg_2
Nokia
67
SvUHeoFtSKw_0
2
68
SvUHeoFtSKw_1
Smooth user experience; 3 days battery life; quick fix self-repair; 100% recycled plastic back; AI camera; refined audio.
69
SvUHeoFtSKw_2
There is no way to know based on the video content.
70
SvUHeoFtSKw_3
Samsung
71
y0ozEcgchhw_0
1
72
y0ozEcgchhw_1
The 2nd place
73
y0ozEcgchhw_2
The Next Applied
74
HtFrFZN8ud4_0
HCL Technologies.
75
HtFrFZN8ud4_1
Mindtree.
76
HtFrFZN8ud4_2
He wears an electronic watch on his left hand.
77
ZuqdsI5hNJ4_0
No. He wears myopia glasses.
78
ZuqdsI5hNJ4_1
He is Li Ka-shing
79
ZuqdsI5hNJ4_2
There are three important reasons for this. 1. Economic growth is now well ahead of the Fed’s long-term potential growth rate. 2. Expansion of the government deficit and investment in clean energy may increase the demand for saving and push up the neutral interest rate. 3. Retirees in industrial economies may be spending those savings now.
80
Y2gLmboNrKk_0
This reveals an important turning point in the global economy from a prolonged low interest rate environment to a possible era of higher interest rates.
81
Y2gLmboNrKk_1
Yes. The dollar bills appear around the 21s of the video.
82
Y2gLmboNrKk_2
Before that, they need to open the vaccine pass and scan the code to enter.
83
wfQlSw2zeZk_0
The reasons for their absence were schedule conflicts or COVID-19 infection, but some media believed that they had received pressure from the US government.
84
wfQlSw2zeZk_1
Her coat is pink.
85
lnShWOBzgGM_0
The Republican Party criticized the Biden administration for waste.
86
lnShWOBzgGM_1
Biden unveils taxes and gets paid less than last year.
87
lnShWOBzgGM_2
The banking sector's share price index rose 2.92%
88
DsCMMJTCGk4_0
Her occupation is a commentator for Securities Daily.
89
DsCMMJTCGk4_1
This video talks about the knowledge of stock trading.
90
akJ6MPa7ujs_0
he was wearing a gray baseball cap.
91
akJ6MPa7ujs_1
26.00
92
akJ6MPa7ujs_2
This video talks about experiences about stock trading.
93
ODzyaMRExpA_0
No. The book is placed under the globe.
94
ODzyaMRExpA_1
He is 46 years old.
95
gNBg3wqo_cM_0
An artist is suspected of having counterfeit NT$500 banknotes on the market. The artist specializes in making counterfeit banknotes and sells them. The studio turns into a counterfeit banknote manufacturing factory.
96
gNBg3wqo_cM_1
Counterfeit banknotes are not saturated in color and smooth to the touch.
97
gNBg3wqo_cM_2
She wears a red coat.
98
YdySNuZSh4Y_0
Interest rates surge, S&P warns of rising debt default risks for Latin American companies
99
YdySNuZSh4Y_1
End of preview.

MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding

Introduction

MMBench-Video is a quantitative benchmark designed to rigorously evaluate LVLMs' proficiency in video understanding. MMBench-Video incorporates approximately 600 web videos with rich context from YouTube, spanning 16 major categories, including News, Sports, etc., covering most video topics people watch in their daily lives. Each video ranges in duration from 30 secs to 6 mins, to accommodate the evaluation of video understanding capabilities on longer videos. The benchmark includes roughly 2,000 original question-answer (QA) pairs, contributed by volunteers, covering a total of 26 fine-grained capabilities. And it also implement a GPT-4-based evaluation paradigm, which offers superior accuracy, consistency, and a closer alignment with human judgments.

Leaderboard

Latest leaderboard is in our openvlm_video_leaderboard.

Data

The dataset includes 1,998 question-answer (QA) pairs, with each QA assessing one or multiple capabilities of a vision-language model. Each question in the dataset is a question-answer questions with groundtruth.

Here is a example:

    index: 177	
    video: DmUgQzu3Z4U	
    video_type: Food & Drink	
    question: Did the mint-style guy in the video drink his mouthwash?	
    answer: Yes, he drank it. This is very strange. Under normal circumstances we are not allowed to drink mouthwash, but this boy may be doing it to attract viewers.	
    dimensions: ['Counterfactual Reasoning']	
    video_path: ./video/DmUgQzu3Z4U.mp4

How to get video data

Using this function to unwrap pkl files to get original video data.

def unwrap_hf_pkl(pth, suffix='.mp4'):
    base_dir = os.path.join(pth, 'video_pkl/')
    target_dir = os.path.join(pth, 'video/')
    pickle_files = [os.path.join(base_dir, file) for file in os.listdir(base_dir)]
    pickle_files.sort()

    if not os.path.exists(target_dir):
        os.makedirs(target_dir, exist_ok=True)
        for pickle_file in pickle_files:
            with open(pickle_file, 'rb') as file:
                video_data = pickle.load(file)
            # For each video file in the pickle file, write its contents to a new mp4 file
            for video_name, video_content in video_data.items():
                output_path = os.path.join(target_dir, f'{video_name}{suffix}')
                with open(output_path, 'wb') as output_file:
                    output_file.write(video_content)
        print('The video file has been restored and stored from the pickle file.')
    else:
        print('The video file already exists.')

For full dataset evaluation, you can use VLMEvalKit to use MMBench-Video with single command.

python run.py --model GPT4o --data MMBench-Video --nframe 8 --verbose

Citation

@misc{fang2024mmbenchvideolongformmultishotbenchmark,
      title={MMBench-Video: A Long-Form Multi-Shot Benchmark for Holistic Video Understanding}, 
      author={Xinyu Fang and Kangrui Mao and Haodong Duan and Xiangyu Zhao and Yining Li and Dahua Lin and Kai Chen},
      year={2024},
      eprint={2406.14515},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2406.14515}, 
}

If you using VLMEvalKit for model evaluation, please cite this:

@misc{duan2024vlmevalkitopensourcetoolkitevaluating,
      title={VLMEvalKit: An Open-Source Toolkit for Evaluating Large Multi-Modality Models}, 
      author={Haodong Duan and Junming Yang and Yuxuan Qiao and Xinyu Fang and Lin Chen and Yuan Liu and Amit Agarwal and Zhe Chen and Mo Li and Yubo Ma and Hailong Sun and Xiangyu Zhao and Junbo Cui and Xiaoyi Dong and Yuhang Zang and Pan Zhang and Jiaqi Wang and Dahua Lin and Kai Chen},
      year={2024},
      eprint={2407.11691},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2407.11691}, 
}

License

The MMBench-Video dataset is licensed under a Creative Commons Attribution 4.0 International License.

Downloads last month
370

Space using opencompass/MMBench-Video 1