xdecoder/Instruct-X-Decoder

1# --------------------------------------------------------2# X-Decoder -- Generalized Decoding for Pixel, Image, and Language3# Copyright (c) 2022 Microsoft4# Licensed under The MIT License [see LICENSE for details]5# Written by Xueyan Zou (xueyan@cs.wisc.edu)6# --------------------------------------------------------7 8import torch9import numpy as np10from PIL import Image11from torchvision import transforms12from utils.visualizer import Visualizer13from detectron2.utils.colormap import random_color14from detectron2.data import MetadataCatalog15from detectron2.structures import BitMasks16 17 18t = []19t.append(transforms.Resize(512, interpolation=Image.BICUBIC))20transform = transforms.Compose(t)21metadata = MetadataCatalog.get('ade20k_panoptic_train')22 23def open_instseg(model, image, texts, inpainting_text, *args, **kwargs):24    thing_classes = [x.strip() for x in texts.split(',')]25    thing_colors = [random_color(rgb=True, maximum=255).astype(np.int32).tolist() for _ in range(len(thing_classes))]26    thing_dataset_id_to_contiguous_id = {x:x for x in range(len(thing_classes))}27 28    MetadataCatalog.get("demo").set(29        thing_colors=thing_colors,30        thing_classes=thing_classes,31        thing_dataset_id_to_contiguous_id=thing_dataset_id_to_contiguous_id,32    )33 34    with torch.no_grad():35        model.model.sem_seg_head.predictor.lang_encoder.get_text_embeddings(thing_classes + ["background"], is_eval=True)36 37        metadata = MetadataCatalog.get('demo')38        model.model.metadata = metadata39        model.model.sem_seg_head.num_classes = len(thing_classes)40 41        image_ori = transform(image)42        width = image_ori.size[0]43        height = image_ori.size[1]44        image = np.asarray(image_ori)45        images = torch.from_numpy(image.copy()).permute(2,0,1).cuda()46 47        batch_inputs = [{'image': images, 'height': height, 'width': width}]48        outputs = model.forward(batch_inputs)49        visual = Visualizer(image_ori, metadata=metadata)50 51        inst_seg = outputs[-1]['instances']52        inst_seg.pred_masks = inst_seg.pred_masks.cpu()53        inst_seg.pred_boxes = BitMasks(inst_seg.pred_masks > 0).get_bounding_boxes()54        demo = visual.draw_instance_predictions(inst_seg) # rgb Image55        res = demo.get_image()56 57 58    MetadataCatalog.remove('demo')59    torch.cuda.empty_cache()60    return Image.fromarray(res), '', None61