plnguyen2908/AudioVisual-Benchmark-Evaluation

AudioVisual Benchmark Evaluation — evaluation subsets

Item-id lists for the audio-visual benchmark subsets used in our reported evaluation tables.

Layout

<benchmark>/eval_subset.csv   item ids evaluated in the paper
<benchmark>/media_index.csv   id -> media filename(s)
<benchmark>/media/            the media files those ids refer to

eval_subset.csv holds a single id column keyed to the source benchmark (question_id, idx, or index). media/ contains exactly the files those ids reference and nothing else. Questions and answers are not redistributed — take them from the source benchmark and join on the id.

Item count and file count differ by design: several benchmarks ask multiple questions about one clip (AVHBench 4803 items / 2021 videos), while others need more than one file per item (OmniBench is one image plus one audio each). media_index.csv gives the exact mapping.

Benchmarks

folder	benchmark	eval subset n	media files	id column
avhbench	AVHBench	4803	4042	question_id
cmm_full	CMM	2316	3463	idx
avs	AV-SpeakerBench	1927	1365	question_id
omnibench	OmniBench	1118	2236	index
videomme	Video-MME (short+medium)	1749	600	question_id
worldsense	WorldSense	3118	1645	index

Evaluation subsets

As mentioned in the paper, we evaluate on the short and medium length subset of all of the benchmarks because we believe that it takes a different approach for long video.

Source benchmarks

benchmark	source
AVHBench	Sung-Bin et al., AVHBench
CMM	Curse of Multi-Modalities
AV-SpeakerBench	plnguyen2908/AV-SpeakerBench
OmniBench	m-a-p/OmniBench
Video-MME	lmms-lab/Video-MME
WorldSense	honglyhly/WorldSense

Refer to each source benchmark for its own license and terms.