Spaces:
Running on Zero
Running on Zero
| class EnvironmentSettings: | |
| def __init__(self): | |
| self.workspace_dir = 'checkpoints' # Base directory for saving network checkpoints. | |
| self.tensorboard_dir = self.workspace_dir # Directory for tensorboard files. | |
| self.pretrained_networks = self.workspace_dir | |
| self.pre_trained_models_dir = self.workspace_dir+"/backup" | |
| ######################################################################################## | |
| self.eval_dataset_name = 'docunet' | |
| if self.eval_dataset_name == 'dir300': | |
| self.eval_dataset = '/home/share/dir300' | |
| elif self.eval_dataset_name == 'docunet': | |
| self.eval_dataset = '/Data_PHD_Backup/phd23_weiguang_zhang/dataset/docker_usecase/shared_data/docunet' | |
| elif self.eval_dataset_name == 'anyphoto': | |
| self.eval_dataset = '/home/share/init_all_final/init_8' | |
| elif self.eval_dataset_name == 'docreal': | |
| self.eval_dataset = '/home/share/docreal' | |
| ######################################################################################## | |
| self.dataset_name = 'doc3d' | |
| if self.dataset_name == 'doc_debug': | |
| self.doc_debug = '/home/share/train_bug3' | |
| self.time_variant = False | |
| elif self.dataset_name == 'aug_doc3d': | |
| self.doc_debug = '/home/share/train_bug3' | |
| self.time_variant = "new" | |
| elif self.dataset_name == 'doc3d': | |
| self.doc_debug = '/home/share/doc3d_rearrange2' | |
| self.time_variant = True | |
| self.train_mode = 'stage_1_dit_cross' | |
| self.iter = True | |
| self.train_VGG = True | |
| self.use_gt_mask = False | |
| self.use_line_mask = True | |
| self.use_init_flow = False | |
| self.lr = 1e-4 | |
| self.diffusion_steps = 3 | |
| self.batch_size = 10 | |
| self.n_threads = 4 | |
| ###################################### | |
| self.log_interval = 20 | |
| self.save_interval = 4000 | |
| self.resume_step = 0 #152000 #1390000 | |
| self.resume_checkpoint = None | |
| self.nbr_objects = 4 | |
| self.min_area_objects = 1300 | |
| self.compute_object_reprojection_mask = True | |
| self.initial_pretrained_model = None | |
| self.data_dir = '' | |
| self.schedule_sampler = 'uniform' #'uniform' 'multi' 'fixed' | |
| self.weight_decay = 0.0 | |
| self.lr_anneal_steps = 0 | |
| self.microbatch = -1 | |
| self.ema_rate = 0.9999 | |
| self.use_fp16 = False | |
| self.fp16_scale_growth = 0.001 | |
| self.image_size = 64 | |
| self.flow_size = (64, 64) | |
| self.num_channels = 128 | |
| self.num_res_blocks = 3 | |
| self.num_heads = 4 | |
| self.num_heads_upsample = -1 | |
| self.attention_resolutions = "16,8" | |
| self.dropout = 0.0 | |
| self.learn_sigma = False | |
| self.sigma_small = False | |
| self.class_cond = False | |
| self.noise_schedule = 'cosine' | |
| self.use_kl = False | |
| self.predict_xstart = True | |
| self.rescale_timesteps = True | |
| self.rescale_learned_sigmas = True | |
| self.use_checkpoint = False | |
| self.use_scale_shift_norm = True | |
| self.clip_denoised = False | |
| self.num_samples = 10000 | |
| self.val_batch_size = 1 | |
| self.use_ddim = False | |
| self.model_path = '/Data_PHD/phd23_weiguang_zhang/project/huggingface_dvd/DvD/checkpoints/model1852000.pt' # 0428_1 99 | |
| self.seg_model_path = "/Data_PHD/phd23_weiguang_zhang/project/huggingface_dvd/DvD/checkpoints/seg.pth" | |
| self.line_seg_model_path = '/Data_PHD/phd23_weiguang_zhang/project/huggingface_dvd/DvD/checkpoints/line_model2.pth' # 'checkpoints/backup/line_model2.pth' 'checkpoints/backup/30.pt' | |
| self.new_seg_model_path = '/Data_PHD/phd23_weiguang_zhang/project/huggingface_dvd/DvD/checkpoints/seg_model.pth' | |
| self.timestep_respacing = '' | |
| self.n_batch = 2 # The number of multiple hypotheses | |
| self.visualize = True # Set True, if you want qualitative results. | |
| self.use_sr_net = False | |