Making Art with GANs : Training Pix2PixHD
Class Recording https://www.youtube.com/watch?v=tSbYOHKue_E
Notebook : https://colab.research.google.com/github/dvschultz/ai/blob/master/NFP_Pix2PixHD.ipynb
Epoch : Files will be save when the model reach the amount of epochs requested
Continue training
!python train_video.py \ --name water \ --dataroot ./datasets/water/ \ --save_epoch_freq 5 \ --niter 150 \ --niter_decay 150 \ --continue_train
niter = number of iterations
Output video should under Checkpoints
Final Product
Notebook https://github.com/dvschultz/ai/blob/master/NFP_Pix2PixHD.ipynb
If running out of space under checkpoint folder
Delete _D (Discriminator)
****START FROM CELL
--DIFFERENT START TIME
Switch the epoch to generate a video from a different start time.
CHANGE IMAGE ON GENERATE
To change the image used in the generator, it can be added on Colab .
***The image added need to be the same size as the video
!python generate_video.py \ --name water \ --dataroot ./datasets/water/ \ --fps 24 \ --how_many 200 \ --which_epoch 110 \ --no_crop \ --start_from /content/NAMEOFIMAGE \ --start_frames_before 0
VIDEO
FOR ZOOM EFFECT
!python generate_video.py \ --name water \ --dataroot ./datasets/water/ \ --fps 24 \ --how_many 200 --which_epoch 110 --zoom_lvl 20 --no_crop