Skip to content

Latest commit

 

History

History
13 lines (13 loc) · 1.02 KB

README.md

File metadata and controls

13 lines (13 loc) · 1.02 KB
Raw

Training

We build an editing prediction model to predict whether a query video contains copied segments and what type of manipulations are used. We first create a simulated copy video dataset using the training reference of the VSC2022 data. Then we train a prediction model using this dataset.

Generate simulated dataset

  1. Download The training set of VSC2022.
  2. Set the path of input video data_path, metadata meta_path, and output folder output. Run the following will generate samples with index from 0 to 4999.
python video_generation_AugLy.py --sample_range 0 5000
  1. The generated data has the following components:
    • copied videos(query/Q4XXXXX.mp4)
    • ground-truth (cvl_train_matching_ground_truth.csv): query_id, ref_id, query_start/end, ref_start/end
    • metadata (cvl_train_query_metadata.csv): video_id duration_sec frames_per_sec width height rn base_query_id
    • manipulation type (cvl_train_augly_metadata/Q4XXXXX.json)