The feature of CycleGAN is that it does not require training data to be paired, and can learn image conversion between different domains. The streets applied to semantic tags are converted into realistic images, satellite images are converted into map images, and the scene is converted from day to night.
[Instructions]
1_visdom_server.ipynb
Open visdom_server with port 8801.
2_visdom_server_browser.ipynb
Open the browser, you can see the loss curve and other information of the training process.
3_train_GPU.ipynb
Train the model.
• --continue_train : Load model/latest model to continue training.
• --n_epochs 1000 : The training epochs is 1000.
• --gpu_ids 0 : Which gpu to use for training, you can also set more than one, for example --gpu_ids 0,1,2.
• --save_epoch_freq 5 : Store model frequency.
4_inference_folder_GPU.ipynb
Inference folder.
• checkpoints_dir = 'model' : Model folder.
• epoch = 'latest', model_suffix = '_A' : Load the latest_net_G_A model.
• dataroot = 'data/testA' : Test folder.
This SDK is built in AppForAI - AI Dev Tools.
Purchase license separately: USD 600, permanent authorization, single APP authorization, single machine authorization, one-year activation, one-year download, one-year update, one-year email technical support.