import os import json from models.segmentor.dynamicUNet import Unet from dataset.transformer import TransformerVal from dataset.dataset import OralSlide, collate from helper.helper_unet import SlideInference, create_model_load_weights from helper.runner import Runner from configs.config_patch_merge_unet import Config distributed = False cfg = Config(mode='patch-merge', train=False) model = Unet(classes=cfg.n_class, encoder_name=cfg.encoder, **cfg.model_cfg) runner = Runner(cfg, model, create_model_load_weights, distributed=distributed) ################################### print("preparing datasets......") slideset_cfg = cfg.testset_cfg slide_list = sorted(os.listdir(slideset_cfg["img_dir"])) transformer = TransformerVal() dataset = OralSlide( slide_list, slideset_cfg["img_dir"], slideset_cfg["meta_file"], slide_mask_dir=slideset_cfg["mask_dir"], label=slideset_cfg['label'], transform=transformer, ) runner.eval_slide(dataset, SlideInference, cfg.test_output_path)
from utils.seg_loss import FocalLoss, SymmetricCrossEntropyLoss, DecoupledSegLoss_v1, DecoupledSegLoss_v2 from helper.helper_unet import Trainer, Evaluator, SlideInference, save_ckpt_model, update_log, update_writer, get_optimizer, create_model_load_weights from configs.config_local_merge_unet import Config from helper.runner import argParser, seed_everything, Runner args = argParser() distributed = False # if torch.cuda.device_count() > 1: # distributed = True # seed SEED = 233 seed_everything(SEED) cfg = Config(mode='local_merge', train=True) model = Unet(classes=cfg.n_class, encoder_name=cfg.encoder, **cfg.model_cfg) runner = Runner(cfg, model, create_model_load_weights, distributed=distributed) ################################### print("preparing datasets......") batch_size = cfg.batch_size num_workers = cfg.num_workers trainset_cfg = cfg.trainset_cfg valset_cfg = cfg.valset_cfg testset_cfg = cfg.testset_cfg transformer_train = TransformerMerge() dataset_train = OralDatasetLocal( trainset_cfg["img_dir"], trainset_cfg["mask_dir"], trainset_cfg["meta_file"], label=trainset_cfg["label"],
from helper.runner import Runner if __name__ == '__main__': Runner().start()
from utils.seg_loss import FocalLoss, SymmetricCrossEntropyLoss, DecoupledSegLoss_v1, DecoupledSegLoss_v2 from helper.helper_unet import Trainer, Evaluator, save_ckpt_model, update_log, update_writer, get_optimizer, create_model_load_weights from configs.config_global_unet import Config from helper.runner import argParser, seed_everything, Runner args = argParser() distributed = False # if torch.cuda.device_count() > 1: # distributed = True # seed SEED = 23 seed_everything(SEED) cfg = Config(mode='global', train=True) model = Unet(classes=cfg.n_class, encoder_name=cfg.encoder, **cfg.model_cfg) runner = Runner(cfg, model, create_model_load_weights, distributed=distributed) ################################### print("preparing datasets......") batch_size = cfg.batch_size num_workers = cfg.num_workers trainset_cfg = cfg.trainset_cfg valset_cfg = cfg.valset_cfg transformer_train = Transformer() dataset_train = OralDataset( trainset_cfg["img_dir"], trainset_cfg["mask_dir"], trainset_cfg["meta_file"], label=trainset_cfg["label"], transform=transformer_train,
import os import json from models.segmentor.dynamicUNet import Unet from dataset.transformer import TransformerVal from dataset.dataset import OralSlide, collate from helper.helper_unet import SlideInference, create_model_load_weights from helper.runner import Runner from configs.config_patch_unet import Config distributed = False cfg = Config(mode='patch', train=False) model = Unet(classes=cfg.n_class, encoder_name=cfg.encoder, **cfg.model_cfg) runner = Runner(cfg, model, create_model_load_weights, distributed=distributed) ################################### print("preparing datasets......") with open('/media/ldy/7E1CA94545711AE6/OSCC/train_val_part.json', 'r') as f: slide_list = json.load(f)['val'] slideset_cfg = cfg.slideset_cfg transformer = TransformerVal() dataset = OralSlide( slide_list, slideset_cfg["img_dir"], slideset_cfg["meta_file"], slide_mask_dir=slideset_cfg["mask_dir"], label=slideset_cfg['label'], transform=transformer, ) runner.eval_slide(dataset, SlideInference)