search

RetroCirce / HTS-Audio-Transformer

The official code repo of "HTS-AT: A Hierarchical Token-Semantic Audio Transformer for Sound Classification and Detection"
https://arxiv.org/abs/2202.00874
MIT License
341 stars 62 forks source link

Audioset dataset for pretraining #38

Open youngwhite opened 1 year ago

youngwhite commented 1 year ago

Dear author, since the model depends on pretraining on AudioSet to reach the highest score, why not to share the dataset and pretrained-model file? For the audioset dataset always become partially invalid.

RetroCirce commented 1 year ago

Hi, our pretrained checkpoint is released, please check our readme released link.

For AudioSet, you can refer to this repo, we use their stored AudioSet (Please check the refered repo's readme 1. Download the dataset) , currently many files in AudioSet in no longer valided. And we save one record at Baidu Pan.

mayqinxu commented 7 months ago

Hi, our pretrained checkpoint is released, please check our readme released link.

For AudioSet, you can refer to this repo, we use their stored AudioSet (Please check the refered repo's readme 1. Download the dataset) , currently many files in AudioSet in no longer valided. And we save one record at Baidu Pan.

Hi RetroCirce, I opened the checkpoint url and felt a little bit confused. What is the difference between the 5 ckpt file under the AudioSet folder?

mayqinxu commented 7 months ago

Hi, our pretrained checkpoint is released, please check our readme released link. For AudioSet, you can refer to this repo, we use their stored AudioSet (Please check the refered repo's readme 1. Download the dataset) , currently many files in AudioSet in no longer valided. And we save one record at Baidu Pan.

Hi RetroCirce, I opened the checkpoint url and felt a little bit confused. What is the difference between the 5 ckpt file under the AudioSet folder?

Oh, I found the answer in a closed issue!