pranitchawla/VCDB-Core-AudioVideo

VCDB Core Audio-Video Retrieval

This repository packages synchronized video and extracted audio from the 528-video core set of VCDB as a symmetric video+audio-to-video+audio retrieval task for MTEB/MOEB. The separate 100,000-video background collection is not included.

Terms and provenance

The source dataset is provided by Fudan University for research purposes only. The source authors and Fudan University make no warranties about the dataset, including non-infringement. Users must review and follow the original VCDB terms. The original page is https://fvl.fudan.edu.cn/dataset/vcdb/list.htm and may be intermittently unavailable.

The source videos were collected from YouTube and MetaCafe. Rights in the underlying media remain with their respective rightsholders. This packaging does not grant additional rights beyond the source terms.

Construction

• Raw core videos: 528
• Audio-video corpus items: 527
• Unique packaged audio files (SHA-256): 461
• Videos without an audio stream: 1
• Audio-video queries: 527
• Unique undirected relevant pairs: 5584
• Directed binary qrels: 11168

Stable IDs remain the original topic-relative video paths. The source video without audio, maradona_hand_of_god/de71afa5fe2363589c28021940d2902d8701efdf.flv, is excluded. Qrels involving it are removed, and every retained query has at least one positive. The 66 repeated audio files belong to distinct source videos and are retained because the joint audio-video items remain distinct copy-detection examples.

Relevance is derived from VCDB's human segment-level video copy annotations: multiple annotations for the same source-video pair are collapsed, and both directions are emitted. Audio similarity was not independently judged; audio is an additional signal paired with the human-annotated video content.

Audio extraction

Source codec counts: {"aac": 512, "adpcm_swf": 13, "mp3": 2}. Packaged codec counts: {"aac": 527}.

AAC elementary streams are remuxed without re-encoding into .m4a. The 15 non-AAC tracks are deterministically encoded to AAC at 128 kb/s with one FFmpeg thread, with metadata and chapters removed. Every packaged track is fully decoded during validation.

Re-encoded tracks:

• beckham_70_yard_goal/0056649006c7a250cb4174f1e924768ba2c9d599.flv: adpcm_swf to AAC; output SHA-256 2a36aa3c1c251962605953d1a95a892da5ef1ee7755fade99cd2cb22b81c896d
• beckham_70_yard_goal/97c71f57e83f6ab16906fdd7edc8e3f88d2ceee1.flv: adpcm_swf to AAC; output SHA-256 9961e8047c62cf221867965853e72429c1539d9f1d8c6e1bc36888b2f133e7f8
• bolt_beijing_100m/098df9c477ecf6c4f3818b198490b0258163015d.flv: adpcm_swf to AAC; output SHA-256 e9088ad5bec444b62ed73e4b9296e3abf6457d7bb6edd5124a1badb1843821a7
• bolt_beijing_100m/359b3ef99acac4136096052c6d812a6248811e10.flv: adpcm_swf to AAC; output SHA-256 36385c29993b875ed22ec3de1fd668f3528656a3f4dee65046c355f6e1948dd5
• brazil_vs_brazil_nike_commercial_2012/0cad474aedea70d42f8d8f2a6f9a21e3c40fa40b.flv: adpcm_swf to AAC; output SHA-256 d3f0b03c1d58da9608fbb2438142fe4ca0ffa48ccd2e9be7b7508608dce613ea
• endless_love/3499b9fd4316131ebf0f5d9bbecf6f3f8600c4fb.flv: adpcm_swf to AAC; output SHA-256 23375ae8f8fb9e12bd79cead89027d9cd71b3d39e85b25aeb0714a357ec8a1b1
• endless_love/ef831f0fe7c633be5ed9edd29dd28d6066ea6d88.flv: adpcm_swf to AAC; output SHA-256 22f06767e9989efe9f4b320cf5b25c24daa00328a38fbbcaf89e4be8a539b55b
• infernal_affairs_1/2620a515a196e5d4329f320f478010e578980bee.flv: adpcm_swf to AAC; output SHA-256 9be767b7c3ceb849f07faa56b51d419ac5d4a65b41ecb6fced1b2830551f103f
• infernal_affairs_1/7e4fe3c124a00ce43ce0b5b0aca342d79e3c1ed7.flv: mp3 to AAC; output SHA-256 efa17a9055f1d58fd3e9999aca1367fda43928c6a43321673caa13339e9427f7
• infernal_affairs_1/c70491f4b4d4a8e90561c11ac929edf430e62641.flv: mp3 to AAC; output SHA-256 ae63b285d6da85c8275d4903bbc60a617fa1961b4f4d5aec00756f7bdc1206ec
• saving_private_ryan_omaha_beach/200130e0c39f0239a22ad3f873feac30e88134df.flv: adpcm_swf to AAC; output SHA-256 72f89d6fab4d1d29c9dc4b1a835fa655969e4eee164211a3d858fbda992efa6a
• saving_private_ryan_omaha_beach/31a250020d2cd699bc5fc42897f7522043713cf5.flv: adpcm_swf to AAC; output SHA-256 d1392b4580bf4f3acda3fb6a35c2d9df56aa9793dfdecc16b84407178834dfb8
• saving_private_ryan_omaha_beach/a257de343d6734f0fc478da98cfedf773ea407a4.mp4: adpcm_swf to AAC; output SHA-256 ff9aef20b7d12eeb09f0328f5601e52433f0735ac296af1b29632935744eefe2
• saving_private_ryan_omaha_beach/aa314017a4a18f05208d8702c27d16d57df3f55f.mp4: adpcm_swf to AAC; output SHA-256 9f9901b63ff2396e58472c5d847f1e0e60145a489997444da370168bdda42220
• saving_private_ryan_omaha_beach/aba6851ca7f78f3a83216005bef58766e3dfb26d.flv: adpcm_swf to AAC; output SHA-256 0b799ee33fccbfeac462aa711e1e99bec322a6c5d034af653a1ac9550352d27f

Configurations

• corpus (test): id: string, video: Video, audio: Audio
• queries (test): id: string, video: Video, audio: Audio
• qrels (test): query-id: string, corpus-id: string, score: int32

Source archive SHA-256

• core_dataset.z01: cc4b2f54ee6e13277c7657061c28b101b97f95e5e3897f1246bab0c432d70ddf (730000000 bytes)
• core_dataset.z02: a1b99131f374c841bb79906704e64555119ffa2307bc6915ac5b843dcabdddc3 (730000000 bytes)
• core_dataset.z03: c34b0119bd136d02c9d288e53a8546cfe2f477f52f459fc7a483313ec32ec394 (730000000 bytes)
• core_dataset.z04: 05c591e74e0e51a05703c6c7b93d6e46bf0c4e80e8ab94b0f1a5ceae688046a5 (730000000 bytes)
• core_dataset.z05: fe346c17fb971833b5f4dbef8c3313d84e47d8d0f6c00f5165b7be6ca65bbf3e (730000000 bytes)
• core_dataset.z06: 01973a3b5eab4ef8451ac6f2434a71d14aff8ebd2c1b0af078dbfd20672cf522 (730000000 bytes)
• core_dataset.z07: 8f700e059bf29b1dfdde4cb9bc6c38b1ad1ba12152c70617a2116e732eb66cb6 (730000000 bytes)
• core_dataset.z08: 6bb5286b05987de11a076b4fd0985017f4ca931bc53cc1cd16510d802690f0a3 (730000000 bytes)
• core_dataset.z09: 84ba7436187dc4bc41ce21475b3f70b5887dd800da8e61beea660650dd322d76 (730000000 bytes)
• core_dataset.z10: 31e9710787b5994118304a9f336312943c81b643a56e5e2425577755d7362f7c (730000000 bytes)
• core_dataset.z11: 7164ad7d655398e092580fdcda734dbead0960e8dee858a62e8dd4218bbc05c4 (730000000 bytes)
• core_dataset.zip: 22f2e052e835f30e4bd1cc794ff2c1a564923fed7d5fa96d8db5afca715a5a16 (260421905 bytes)

Limitations

VCDB was designed for temporal partial-copy detection in video, and the audio signal may have been modified or replaced in copied segments. This adaptation evaluates whole audio-video items, uses only the core set, and cannot reproduce the paper's segment-level F1 score or complete 100K-background protocol.

Citation

@inproceedings{jiang2014vcdb,
  author = {Jiang, Yu-Gang and Jiang, Yudong and Wang, Jiajun},
  booktitle = {Computer Vision -- ECCV 2014},
  pages = {357--371},
  publisher = {Springer},
  title = {VCDB: A Large-Scale Database for Partial Copy Detection in Videos},
  year = {2014},
}

Paper: https://fvl.fudan.edu.cn/_upload/article/files/7b/c4/190424104d2192e8e83cb9dfa6fc/5a8d85f7-5738-450c-b957-3de71d8d7e72.pdf