wentaozhu/AutoShot

Inference on video

Opened this issue · 9 comments

Vikesh-04 commented

Hi, Thanks for the amazing work. I was wondering if there is a way to test it on user-passed video as input. Please highlight the steps to run inference.
Thanks.

danielcrane commented

I'd also be interested in this, I tried to use the existing code to run one of my own videos, but the results were basically noise - hopefully I just did something wrong!

eisneim commented

@danielcrane just use transnet v2, it's very easy to use the pytorch version

  • 👍3
danielcrane commented

@danielcrane just use transnet v2, it's very easy to use the pytorch version

@eisneim Indeed yes I'm currently making use of that, it's very good, just wanted to test performance differences!

  • 👍1
lrbk commented

@danielcrane just use transnet v2, it's very easy to use the pytorch version

@eisneim Indeed yes I'm currently making use of that, it's very good, just wanted to test performance differences!

@eisneim ,well I use transnet v2 and autoshot both for my own dataset , but I find noise and misjudge for gradual transition ,do you know how to fix it?

eisneim commented

@lrbk for long transitions i found no solutions, so what i did is cut them manually; but for the most part TransnetV2 is saving me a lot time

lrbk commented

@lrbk for long transitions i found no solutions, so what i did is cut them manually; but for the most part TransnetV2 is saving me a lot time

@eisneim I follow the training instruction of TransNetV2, but i found some problem with gin and its config parameters.Could you tell me which version of gin and other library you used?

eisneim commented

@lrbk i didn't do any training, just use the pretrained weights

runrunaway2020 commented

@lrbk for long transitions i found no solutions, so what i did is cut them manually; but for the most part TransnetV2 is saving me a lot time

@eisneim I follow the training instruction of TransNetV2, but i found some problem with gin and its config parameters.Could you tell me which version of gin and other library you used?

Hello, I am also trying to copy the training of transnetv2. Have you succeeded? Are there any special configurations that need to be paid attention to?

eisneim commented

i didn't do any training, just use the pretrained weights

i didn't do any training, just use the pretrained weights to find cut point and divide the video into small parts