轻量化的diffusion模型。 Lightweight diffusion model.
Install project
conda create -n diffusion python 3.x
conda activate diffusion
cd basics_diffusion
pip install -r requirements.txtEnter the config file and modify it.进入config文件下进行修改。
Num_epochs = 50000 训练次数。Training times.
Image_size = 64 输入或生成图片大小。Enter or generate picture size.
Dataset_root = "F:/datasets/anime_1024x1024_tiny/images" 数据集地址，文件夹包含图片。Data set address, subfolder containing pictures.
Batch_size = 30 数据集的训练批次。Training batch of data set.
Noise_timesteps = 1000 生成噪声的步数，越短训练越快。The shorter the steps to generate noise, the faster the training.
Weights_path = "weights/model_64x64.pth" 设置权重生成的地址和名称。Set the address and name of the weight generation.
Learning_rate = 5e-5 学习率。配置完config.py文件，再执行文件train.py。 After configuring the config.py file, execute the file .
python train.py在test()函数下，gen_image_num是选择生成几张图片。Under the test() function, gen_ image_ Num is the option to generate several pictures.
python test.py