import os os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE" import torch from annotator.oneformer.detectron2.config import get_cfg from annotator.oneformer.detectron2.projects.deeplab import add_deeplab_config from annotator.oneformer.detectron2.data import MetadataCatalog from annotator.oneformer.oneformer import ( add_oneformer_config, add_common_config, add_swin_config, add_dinat_config, ) from annotator.oneformer.oneformer.demo.defaults import DefaultPredictor from annotator.oneformer.oneformer.demo.visualizer import Visualizer, ColorMode def make_detectron2_model(config_path, ckpt_path): cfg = get_cfg() add_deeplab_config(cfg) add_common_config(cfg) add_swin_config(cfg) add_oneformer_config(cfg) add_dinat_config(cfg) cfg.merge_from_file(config_path) cfg.MODEL.WEIGHTS = ckpt_path cfg.freeze() metadata = MetadataCatalog.get(cfg.DATASETS.TEST_PANOPTIC[0] if len(cfg.DATASETS.TEST_PANOPTIC) else "__unused") return DefaultPredictor(cfg), metadata def semantic_run(img, predictor, metadata): predictions = predictor(img[:, :, ::-1], "semantic") # Predictor of OneFormer must use BGR image !!! visualizer_map = Visualizer(img, is_img=False, metadata=metadata, instance_mode=ColorMode.IMAGE) out_map = visualizer_map.draw_sem_seg(predictions["sem_seg"].argmax(dim=0).cpu(), alpha=1, is_text=False).get_image() return out_map