Evaluation Framework for CCMusic Database Classification Tasks

Classify spectrograms by fine-tuned pre-trained CNN models.

Download

git clone git@github.com:monetjoe/ccmusic_eval.git
cd ccmusic_eval

Requirements

Conda + Pip

conda create -n cv --yes --file conda.txt -c nvidia
conda activate cv
pip install -r requirements.txt

Pip only

pip install torch torchvision torchaudio -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

Usage

python train.py --ds ccmusic-database/chest_falsetto --subset eval --data cqt --label singing_method --model squeezenet1_1 --fl True --mode 0

Help

Args	Notes	Options	Type
--ds	The dataset on ModelScope to be evaluated	For examples: ccmusic-database/chest_falsetto, ccmusic-database/bel_canto	string
--subset	The subset of the dataset	For examples: default, eval	string
--data	Input data colum of the dataset	For examples: mel, cqt, chroma	string
--label	Label colum of the dataset	For examples: label, singing_method, gender	string
--model	Select a CV backbone to train	Supported backbones	string
--imgnet	ImageNet version the backbone was pretrained on	v1, v2	string
--mode	Training mode	2=full_finetune, 1=linear_probe, 0=no_pretrain	int
--bsz	Batch size	For examples: 1, 2, 4, 8, 16, 32, 64, 128..., default is 4	int
--eps	Epoch number	Default is 40	int
--fl	Whether to use focal loss	True, False	bool

Fixed Hyper Params

Param	Value	Range
iteration	10	train
lr	0.001	optimizer
momentum	0.9	optimizer
optimizer	SGD	scheduler
mode	min	scheduler
factor	0.1	scheduler
patience	5	scheduler
verbose	True	scheduler
threshold	lr	scheduler
threshold_mode	rel	scheduler
cooldown	0	scheduler
min_lr	0	scheduler
eps	1e-08	scheduler

Cite

@dataset{zhaorui_liu_2021_5676893,
  author       = {Monan Zhou, Shenyang Xu, Zhaorui Liu, Zhaowen Wang, Feng Yu, Wei Li and Baoqiang Han},
  title        = {CCMusic: an Open and Diverse Database for Chinese and General Music Information Retrieval Research},
  month        = {mar},
  year         = {2024},
  publisher    = {HuggingFace},
  version      = {1.2},
  url          = {https://huggingface.co/ccmusic-database}
}