xinhaomei / act Goto Github PK

Hi thanks for the code. I want to use the pretrained model for making inference on my own data. So the data is only audio files (and no csv files for training). I set up the environment, then put data files in data folder (waveforms). Updated settings.yaml file as instructed (eval mode, pretrained model etc.). Could you please guide me about making inference now? (I tried running train.py but it gives errors about missing train.h5 file etc.). Thanks.

Hello, Dr. Mei! Thank you for open-sourcing this awesome work.

While the procedure to reproduce ACT_m_DeiT and ACT_m_scratch was sufficiently explained, I was wondering if it would be possible to upload them / otherwise obtain from you the model you used to produce the Results Table - Just like you did with the ACT_*_AudioSet_DeiT series.

Thank you!

Hi,
can you provide another way for us to download audiocaps data? I can't use Baidu.

Could i download word_list.p ?

Hello, nice work!
After I replicate your work, and I got the model trained from scratch by myself ( the encoder used pretrained_model -- audioset_deit).
And when I want to evaluate the model, I set the config.path.eval_model to the path where my model is, but when I load the state_dict, there is something wrong.

I do not know why there is nothing about decoder? Could you help me with this problem? Thank you for your early reply!!!

Hi, thank you for this valuable resource.
The train.zip provided on the google drive, upon unzipping gives the following error message:
file #9024: bad zipfile offset (local header sig): 4819319277
error: invalid zip file with overlapped components (possible zip bomb)

How shall I resolve this issue?