-
Notifications
You must be signed in to change notification settings - Fork 7
/
runme.sh
40 lines (28 loc) · 2.59 KB
/
runme.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
#!/bin/bash
DCASE2018_TASK1_DATASET_DIR="/vol/vssp/datasets/audio/dcase2018/task1/TUT-urban-acoustic-scenes-2018-development"
DCASE2018_TASK2_DATASET_DIR="/vol/vssp/datasets/audio/dcase2018/task2"
WORKSPACE="/vol/vssp/msos/qk/workspaces/weak_source_separation/dcase2018_task2"
# Create DCASE 2018 Task 2 cross-validation csv. Only manually verified sound events are used for cross validation.
python utils/create_mixture_yaml.py create_dcase2018_task2_cross_validation_csv --dcase2018_task2_dataset_dir=$DCASE2018_TASK2_DATASET_DIR --workspace=$WORKSPACE
# Create mixture yaml file of sound events and background noise.
python utils/create_mixture_yaml.py create_mixture_yaml --dcase2018_task1_dataset_dir=$DCASE2018_TASK1_DATASET_DIR --dcase2018_task2_dataset_dir=$DCASE2018_TASK2_DATASET_DIR --workspace=$WORKSPACE
# Create mixed audios
SNR=0
python utils/create_mixed_audio.py --dcase2018_task1_dataset_dir=$DCASE2018_TASK1_DATASET_DIR --dcase2018_task2_dataset_dir=$DCASE2018_TASK2_DATASET_DIR --workspace=$WORKSPACE --scene_type=dcase2018_task1 --snr=$SNR
# Calculate features
python utils/features.py logmel --workspace=$WORKSPACE --scene_type=dcase2018_task1 --snr=$SNR
# Train
MODEL_TYPE="gmp" # 'gmp' | 'gap' | 'gwrp'
HOLDOUT_FOLD=1
CUDA_VISIBLE_DEVICES=1 python pytorch/main_pytorch.py train --workspace=$WORKSPACE --model_type=$MODEL_TYPE --scene_type=dcase2018_task1 --snr=$SNR --holdout_fold=$HOLDOUT_FOLD --cuda
# Inference
CUDA_VISIBLE_DEVICES=1 python pytorch/main_pytorch.py inference --workspace=$WORKSPACE --model_type=$MODEL_TYPE --scene_type=dcase2018_task1 --snr=$SNR --holdout_fold=$HOLDOUT_FOLD --iteration=10000 --cuda
# Get average statistics
python utils/get_avg_stats.py single_fold --workspace=$WORKSPACE --filename=main_pytorch --model_type=$MODEL_TYPE --scene_type=dcase2018_task1 --snr=$SNR --holdout_fold=$HOLDOUT_FOLD
# After train & inference and calculate the statistics of folds 1, 2, 3 and 4, you may run the following command to get averaged statistics of all folds.
python utils/get_avg_stats.py all_fold --workspace=$WORKSPACE --filename=main_pytorch --model_type=$MODEL_TYPE --scene_type=dcase2018_task1 --snr=$SNR
############# Plot figures for paper #############
# Visualize waveform & spectrogram
python utils/visualize.py waveform --workspace=$WORKSPACE --scene_type=dcase2018_task1 --snr=$SNR --holdout_fold=$HOLDOUT_FOLD --cuda
# Visualze learned segmentation masks & SED results
CUDA_VISIBLE_DEVICES=1 python utils/visualize.py mel_masks --workspace=$WORKSPACE --model_type=$MODEL_TYPE --scene_type=dcase2018_task1 --snr=$SNR --holdout_fold=$HOLDOUT_FOLD --iteration=10000 --cuda