Open andydelworth opened 3 months ago
Specifically, I am looking to create the min.json for this set. I figure I can set most of these fields to be dummy values since I will do the evaluation myself on the inference outputs. But I am wondering how to set the audio_frames field, shown below:
{
"file": "test/000001.mp4",
"n_fakes": 0,
"fake_periods": [],
"duration": 4.224,
"original": null,
"modify_video": false,
"modify_audio": false,
"split": "test",
"video_frames": 103,
"audio_channels": 1,
"audio_frames": 65536
}
Any wisdom is appreciated. Thank you!!
The audio_frames
is the length of the audio waveform.
I'm looking to benchmark your checkpoints on a video dataset that I have created. Do you have advice on the easiest way to do this?
If it's for temporal localization, I suggest modifying the dataloader to fit your dataset. If each video of your dataset is too long to pad, you can resize the video in temporal axis to reduce the length. For classification, you can set the fake_periods
as [[0, duration]]
for training. In test, you can use the boundary map output to thresholding if it is real or fake. Or you can use the auxiliary classification output from self.video_frame_classifiers
and self.audio_frame_classifiers
.
Hi,
Thanks for your awesome work. I'm looking to benchmark your checkpoints on a video dataset that I have created. Do you have advice on the easiest way to do this?
Thanks! Andy