data_preprocess.sh

soundlist=" "dog" "fireworks" "drum" "baby" "gun" "sneeze" "cough" "hammer" "

for soundtype in $soundlist 
do

    # Data preprocessing. We will first pad all videos to 10s and change video FPS and audio
    # sampling rate.
    python extract_audio_and_video.py \
    -i data/VAS/videos/${soundtype} \
    -o data/features/${soundtype}

    # Generating RGB frame and optical flow. This script uses CPU to calculate optical flow,
    # which may take a very long time. We strongly recommend you to refer to TSN repository 
    # (https://github.com/yjxiong/temporal-segment-networks) to speed up this process.
    python extract_rgb_flow.py \
    -i data/features/${soundtype}/videos_10s_21.5fps \
    -o data/features/${soundtype}/OF_10s_21.5fps

    #Split training/testing list
    python gen_list.py \
    -i data/VAS/videos/${soundtype} \
    -o filelists --prefix ${soundtype}

    #Extract Mel-spectrogram from audio
    python extract_mel_spectrogram.py \
    -i data/features/${soundtype}/audio_10s_22050hz \
    -o data/features/${soundtype}/melspec_10s_22050hz

    #Extract RGB feature
    CUDA_VISIBLE_DEVICES=6 python extract_feature.py \
    -t filelists/${soundtype}_train.txt \
    -m RGB \
    -i data/features/${soundtype}/OF_10s_21.5fps \
    -o data/features/${soundtype}/feature_rgb_bninception_dim1024_21.5fps

    CUDA_VISIBLE_DEVICES=6 python extract_feature.py \
    -t filelists/${soundtype}_test.txt \
    -m RGB \
    -i data/features/${soundtype}/OF_10s_21.5fps \
    -o data/features/${soundtype}/feature_rgb_bninception_dim1024_21.5fps

    #Extract optical flow feature
    CUDA_VISIBLE_DEVICES=6 python extract_feature.py \
    -t filelists/${soundtype}_train.txt \
    -m Flow \
    -i data/features/${soundtype}/OF_10s_21.5fps \
    -o data/features/${soundtype}/feature_flow_bninception_dim1024_21.5fps

    CUDA_VISIBLE_DEVICES=6 python extract_feature.py \
    -t filelists/${soundtype}_test.txt \
    -m Flow \
    -i data/features/${soundtype}/OF_10s_21.5fps \
    -o data/features/${soundtype}/feature_flow_bninception_dim1024_21.5fps

done