search

AliaksandrSiarohin / first-order-model

This repository contains the source code for the paper First Order Motion Model for Image Animation
https://aliaksandrsiarohin.github.io/first-order-model-website/
MIT License
14.54k stars 3.22k forks source link

Some more details about training process. #209

Open coinsbarboss opened 4 years ago

coinsbarboss commented 4 years ago

Hi,

I read out the instructions but did not find that one describing the training process for Vox-model.

As I understand I need to create a folder "data/vox". Under that folder, I would create train/ and test/.

Let's say I have a source image - source.png and driving video - driving.mp4

  1. Where should I put those files?
  2. Should I create a new YAML config file or I could use vox-adv-256.yaml ?
  3. Should I create a CSV file for this dataset

I will greatly appreciate the advice!

AliaksandrSiarohin commented 4 years ago

There is no images at training time only videos. So put all the videos to the train folder, if you don't need to compute metrics. If your are using your dataset create new yaml.

coinsbarboss commented 4 years ago

There is no images at training time only videos. So put all the videos to the train folder, if you don't need to compute metrics. If your are using your dataset create new yaml.

Thanks for your fast response.

  1. How should look the new yaml? Should I just copy params from vox-adv-256.yaml and adjust root dir?
  2. If I'm using a new dataset that actually consists of 2 videos ( driving and source ) may I still use the checkpoint?
  3. Should I create a CSV file for this dataset?
AliaksandrSiarohin commented 4 years ago

This is not a new dataset. It is called new sample, use demo.py