gigagan pytorch下載 - gigagan pytorch原始碼下載

我還將添加一些輕量級 gan 的發現，以實現更快的收斂（跳過層激勵）和更好的穩定性（鑑別器中的重建輔助損失）

它還包含 1k - 4k 上採樣器的程式碼，我認為這是本文的亮點。

$ pip install gigagan-pytorch

 import torch

from gigagan_pytorch import (
    GigaGAN ,
    ImageDataset
)

gan = GigaGAN (
    generator = dict (
        dim_capacity = 8 ,
        style_network = dict (
            dim = 64 ,
            depth = 4
        ),
        image_size = 256 ,
        dim_max = 512 ,
        num_skip_layers_excite = 4 ,
        unconditional = True
    ),
    discriminator = dict (
        dim_capacity = 16 ,
        dim_max = 512 ,
        image_size = 256 ,
        num_skip_layers_excite = 4 ,
        unconditional = True
    ),
    amp = True
). cuda ()

# dataset

dataset = ImageDataset (
    folder = '/path/to/your/data' ,
    image_size = 256
)

dataloader = dataset . get_dataloader ( batch_size = 1 )

# you must then set the dataloader for the GAN before training

gan . set_dataloader ( dataloader )

# training the discriminator and generator alternating
# for 100 steps in this example, batch size 1, gradient accumulated 8 times

gan (
    steps = 100 ,
    grad_accum_every = 8
)

# after much training

images = gan . generate ( batch_size = 4 ) # (4, 3, 256, 256)

 import torch
from gigagan_pytorch import (
    GigaGAN ,
    ImageDataset
)

gan = GigaGAN (
    train_upsampler = True ,     # set this to True
    generator = dict (
        style_network = dict (
            dim = 64 ,
            depth = 4
        ),
        dim = 32 ,
        image_size = 256 ,
        input_image_size = 64 ,
        unconditional = True
    ),
    discriminator = dict (
        dim_capacity = 16 ,
        dim_max = 512 ,
        image_size = 256 ,
        num_skip_layers_excite = 4 ,
        multiscale_input_resolutions = ( 128 ,),
        unconditional = True
    ),
    amp = True
). cuda ()

dataset = ImageDataset (
    folder = '/path/to/your/data' ,
    image_size = 256
)

dataloader = dataset . get_dataloader ( batch_size = 1 )

gan . set_dataloader ( dataloader )

# training the discriminator and generator alternating
# for 100 steps in this example, batch size 1, gradient accumulated 8 times

gan (
    steps = 100 ,
    grad_accum_every = 8
)

# after much training

lowres = torch . randn ( 1 , 3 , 64 , 64 ). cuda ()

images = gan . generate ( lowres ) # (1, 3, 256, 256)

健康的運作將使G 、 MSG 、 D 、 MSD值徘徊在0到10之間，並且通常保持相當恆定。如果在 1k 訓練步驟之後的任何時候這些值持續保持在三位數，則表示出現了問題。生成器和鑑別器的值偶爾會下降為負值是可以的，但它應該回升至上述範圍。

GP和SSL應推向0 。 GP偶爾會出現峰值；我喜歡將其想像為網路正在經歷某種頓悟

GigaGAN級現在配備了？加速器。您可以使用其accelerate CLI 輕鬆地透過兩步驟進行多 GPU 訓練

$ accelerate launch train . py

 @misc { https://doi.org/10.48550/arxiv.2303.05511 ,
    url     = { https://arxiv.org/abs/2303.05511 } ,
    author  = { Kang, Minguk and Zhu, Jun-Yan and Zhang, Richard and Park, Jaesik and Shechtman, Eli and Paris, Sylvain and Park, Taesung } ,  
    title   = { Scaling up GANs for Text-to-Image Synthesis } ,
    publisher = { arXiv } ,
    year    = { 2023 } ,
    copyright = { arXiv.org perpetual, non-exclusive license }
}

 @article { Liu2021TowardsFA ,
    title   = { Towards Faster and Stabilized GAN Training for High-fidelity Few-shot Image Synthesis } ,
    author  = { Bingchen Liu and Yizhe Zhu and Kunpeng Song and A. Elgammal } ,
    journal = { ArXiv } ,
    year    = { 2021 } ,
    volume  = { abs/2101.04775 }
}

 @inproceedings { dao2022flashattention ,
    title   = { Flash{A}ttention: Fast and Memory-Efficient Exact Attention with {IO}-Awareness } ,
    author  = { Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R{'e}, Christopher } ,
    booktitle = { Advances in Neural Information Processing Systems } ,
    year    = { 2022 }
}

 @inproceedings { Karras2020ada ,
    title     = { Training Generative Adversarial Networks with Limited Data } ,
    author    = { Tero Karras and Miika Aittala and Janne Hellsten and Samuli Laine and Jaakko Lehtinen and Timo Aila } ,
    booktitle = { Proc. NeurIPS } ,
    year      = { 2020 }
}

 @article { Xu2024VideoGigaGANTD ,
    title   = { VideoGigaGAN: Towards Detail-rich Video Super-Resolution } ,
    author  = { Yiran Xu and Taesung Park and Richard Zhang and Yang Zhou and Eli Shechtman and Feng Liu and Jia-Bin Huang and Difan Liu } ,
    journal = { ArXiv } ,
    year    = { 2024 } ,
    volume  = { abs/2404.12388 } ,
    url     = { https://api.semanticscholar.org/CorpusID:269214195 }
}