Learning to Answer Questions in Dynamic Audio-Visual Scenarios
Paper • 2203.14072 • Published
id stringlengths 6 6 | audio audioduration (s) 33.6 60.9 |
|---|---|
c-0000 | |
c-0001 | |
c-0002 | |
c-0003 | |
c-0004 | |
c-0005 | |
c-0006 | |
c-0007 | |
c-0008 | |
c-0009 | |
c-0010 | |
c-0011 | |
c-0012 | |
c-0013 | |
c-0014 | |
c-0015 | |
c-0016 | |
c-0017 | |
c-0018 | |
c-0019 | |
c-0020 | |
c-0021 | |
c-0022 | |
c-0023 | |
c-0024 | |
c-0025 | |
c-0026 | |
c-0027 | |
c-0028 | |
c-0029 | |
c-0030 | |
c-0031 | |
c-0032 | |
c-0033 | |
c-0034 | |
c-0035 | |
c-0036 | |
c-0037 | |
c-0038 | |
c-0039 | |
c-0040 | |
c-0041 | |
c-0042 | |
c-0043 | |
c-0044 | |
c-0045 | |
c-0046 | |
c-0047 | |
c-0048 | |
c-0049 | |
c-0050 | |
c-0051 | |
c-0052 | |
c-0053 | |
c-0054 | |
c-0055 | |
c-0056 | |
c-0057 | |
c-0058 | |
c-0059 | |
c-0060 | |
c-0061 | |
c-0062 | |
c-0063 | |
c-0064 | |
c-0065 | |
c-0066 | |
c-0067 | |
c-0068 | |
c-0069 | |
c-0070 | |
c-0071 | |
c-0072 | |
c-0073 | |
c-0074 | |
c-0075 | |
c-0076 | |
c-0077 | |
c-0078 | |
c-0079 | |
c-0080 | |
c-0081 | |
c-0082 | |
c-0083 | |
c-0084 | |
c-0085 | |
c-0086 | |
c-0087 | |
c-0088 | |
c-0089 | |
c-0090 | |
c-0091 | |
c-0092 | |
c-0093 | |
c-0094 | |
c-0095 | |
c-0096 | |
c-0097 | |
c-0098 | |
c-0099 |
End of preview. Expand in Data Studio
MusicAVQA-V2A-Retrieval
This is a derived retrieval benchmark from the test split of
mteb/MUSIC-AVQA_cls-preprocessed at
revision 29f50ae80ad4e8c1cfdbc0148aefe6fe050833dd. It uses video queries and audio corpus items.
Construction
The source clips are labelled with 22 musical-instrument classes. For every class, a deterministic seed (42) selects five clips as queries and ten distinct clips as corpus items. Relevance is class membership, so each query has ten relevant corpus items. The resulting test split has 110 queries, 220 corpus items, and 1,100 qrels.
Query and corpus source video_id values are unique and disjoint. This is
therefore cross-clip instrument retrieval, rather than matching a modality to
the synchronous counterpart of its own clip. It evaluates instrument-level
audio-video association and may retain task/scene cues from the source data.
Provenance and license
The source is MUSIC-AVQA from Music Audio-Visual Question Answering (Li et al., 2022). The MUSIC-AVQA project states that its datasets and benchmarks are available under CC BY-NC 4.0. This derived dataset is published under the same non-commercial license.
The exact construction is in
scripts/data/music_avqa_retrieval/create_data.py.
- Downloads last month
- 101