about test dataset

[wclemonade]

hi, i want to know the real test datasets are from 3 datasets (VC1, Zscope, OSora), this means the test is on every fake test datasets of the diffusion generative methods ,and the the above 3 real test datasets?

[SparkleXFantasy]

Hi.

this means the test is on every fake test datasets of the diffusion generative methods ,and the the above 3 real test datasets?

Yes. For fair evaluation, we aggregate these 3 datasets to form an entire real dataset, and combine it with every fake dataset for testing.

[SparkleXFantasy]

Hi! This 'stop_count' parameter is equivalent to 'sample_size' in the bash for test. If you set the value as positive, only the first {stop_count} videos will be evaluated for each class. For precise evaluation, we set the value as '-1' for evaluation on all videos. To save time, we recommend that this parameter can be set as an appropriate positive number (e.g. 500, etc) to get a rough result. I hope this message is helpful to you. 秋枫 ***@***.***

[wclemonade]

Hi! This 'stop_count' parameter is equivalent to 'sample_size' in the bash for test. If you set the value as positive, only the first {stop_count} videos will be evaluated for each class. For precise evaluation, we set the value as '-1' for evaluation on all videos. To save time, we recommend that this parameter can be set as an appropriate positive number (e.g. 500, etc) to get a rough result. I hope this message is helpful to you. 秋枫 @.***

Hi. My device is a single RTX 3090. May I ask how long does it take for you to test a single dataset?

[SparkleXFantasy]

Hi. I think the inference time may be several hours for one single dataset, depending on the number of videos.