Skip to content

Latest commit

 

History

History
74 lines (54 loc) · 4.77 KB

README.md

File metadata and controls

74 lines (54 loc) · 4.77 KB
Raw

README for AudioScope YFCC100m clip lists.

This is a recipe for data based on YFCC100m [1] used in our ICLR 2021 paper [2] on audio-visual on-screen sound separation.

Specifically, we provide CSVs that describe the exact videos and timestamps used for labeled and unlabeled train, validation, and test clips, as well as specification of pairs of clips used to create mixture of mixtures (MoM) validation and test sets. The same train/validation/test splits are used as in these lists. All clips referenced by these lists have been filtered by an unsupervised audio-visual coincidence model trained on AudioSet [3] using a threshold of 0.8 on the predicted coincidence probability.

Split Label Count CSV name
Train None 324970 filtered_train_clips.csv
Train On-screen only 836 filtered_train_onscreen_unanimous_clips.csv
Train Off-screen only 3672* filtered_train_offscreen_unanimous_clips.csv
Validation None 6429 filtered_validate_clips.csv
Validation On-screen only 735 filtered_validate_onscreen_unanimous_clips.csv
Validation Off-screen only 836 filtered_validate_offscreen_unanimous_clips.csv
Test None 3293 filtered_test_clips.csv
Test On-screen only 295 filtered_test_onscreen_unanimous_clips.csv
Test Off-screen only 370 filtered_test_offscreen_unanimous_clips.csv

* The paper [2] gives the incorrect count of 3681.

Addionally, we provide lists of the pairs of clips used to create MoMs for validation and test, where the MoM video uses video frames from the first clip, and a soundtrack that is the sum of the audio from both clips.

Split Label Count CSV name
Validation On-screen + off-screen 3675 filtered_validate_onscreen_unanimous_plus_offscreen_unanimous_mom_clips.csv
Validation Off-screen + off-screen 4180 filtered_validate_offscreen_unanimous_plus_offscreen_unanimous_mom_clips.csv
Test On-screen + off-screen 1475 filtered_test_onscreen_unanimous_plus_offscreen_unanimous_mom_clips.csv
Test Off-screen + off-screen 1850 filtered_test_offscreen_unanimous_plus_offscreen_unanimous_mom_clips.csv

Download instructions

The CSVs are hosted on Google Cloud. They can be downloaded using the following command:

gsutil -m cp -r gs://gresearch/sound_separation/audioscope_yfcc100m_clip_lists .

which will copy the CSVs to the current folder.

CSV format

The train, validation, and test CSVs contain the following columns:

  • Video path: string, path to MP4 video in YFCC100m. Here is an example of a video path: data/videos/mp4/827/f6b/827f6b53467db2d5218ed8247418c4c.mp4
  • Input start time: float, clip start time in seconds.
  • Input end time: float, clip end time in seconds.

The validation and test MoM CSVs contain the following columns, with each row describing two clips that are used to construct the MoM:

  • Video path 1
  • Input start time 1
  • Input end time 1
  • Video path 2
  • Input start time 2
  • Input end time 2

Data License

These lists are released under a Creative Commons license (CC-BY 4.0).

References

[1] Bart Thomee, David Shamma, Gerald Friedland, Benjamin Elizalde, Karl Ni, Dough Poland, Damian Borth, Li-Jia Li, "YFCC100M: The New Data in Multimedia Research", Communications of the ACM, 59(2), pp. 64-73, 2016.

[2] Efthymios Tzinis, Scott Wisdom, Aren Jansen, Shawn Hershey, Tal Remez, Daniel P. W. Ellis, John R. Hershey, "Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds", International Conference on Learning Representations (ICLR), 2021.

[3] Aren Jansen, Daniel P. W. Ellis, Shawn Hershey, R. Channing Moore, Manoj Plakal, Ashok C. Popat, and Rif A. Saurous, "Coincidence, categorization, and consolidation: Learning to recognize sounds with minimal supervision", IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), pp. 121–125, 2020.