import os import sys sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), 'depth'))) sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), 'stable-diffusion'))) sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), 'taming-transformers'))) os.chdir(os.path.abspath(os.path.join(os.path.dirname(__file__), 'depth'))) import cv2 import numpy as np import torch import torch.backends.cudnn as cudnn from depth.models_depth.model import EVPDepth from depth.configs.train_options import TrainOptions from depth.configs.test_options import TestOptions import glob import utils import torchvision.transforms as transforms from utils_depth.misc import colorize from PIL import Image import torch.nn.functional as F import gradio as gr import tempfile css = """ #img-display-container { max-height: 50vh; } #img-display-input { max-height: 40vh; } #img-display-output { max-height: 40vh; } """ def create_demo(model, device): gr.Markdown("### Depth Prediction demo") with gr.Row(): input_image = gr.Image(label="Input Image", type='pil', elem_id='img-display-input') depth_image = gr.Image(label="Depth Map", elem_id='img-display-output') raw_file = gr.File(label="16-bit raw depth, multiplier:256") submit = gr.Button("Submit") def on_submit(image): transform = transforms.ToTensor() image = transform(image).unsqueeze(0).to(device) shape = image.shape image = torch.nn.functional.interpolate(image, (440,480), mode='bilinear', align_corners=True) image = F.pad(image, (0, 0, 40, 0)) with torch.no_grad(): pred = model(image)['pred_d'] pred = pred[:,:,40:,:] pred = torch.nn.functional.interpolate(pred, shape[2:], mode='bilinear', align_corners=True) pred_d_numpy = pred.squeeze().cpu().numpy() colored_depth, _, _ = colorize(pred_d_numpy, cmap='gray_r') tmp = tempfile.NamedTemporaryFile(suffix='.png', delete=False) raw_depth = Image.fromarray((pred_d_numpy*256).astype('uint16')) raw_depth.save(tmp.name) return [colored_depth, tmp.name] submit.click(on_submit, inputs=[input_image], outputs=[depth_image, raw_file]) examples = gr.Examples(examples=["test_img.jpg"], inputs=[input_image]) def main(): opt = TestOptions().initialize() args = opt.parse_args() args.ckpt_dir = 'best_model_nyu.ckpt' device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = EVPDepth(args=args, caption_aggregation=True) cudnn.benchmark = True model.to(device) model_weight = torch.load(args.ckpt_dir, map_location=device)['model'] if 'module' in next(iter(model_weight.items()))[0]: model_weight = OrderedDict((k[7:], v) for k, v in model_weight.items()) model.load_state_dict(model_weight, strict=False) model.eval() title = "# EVP" description = """Official demo for **EVP: Enhanced Visual Perception using Inverse Multi-Attentive Feature Refinement and Regularized Image-Text Alignment**. EVP is a deep learning model for metric depth estimation from a single image. Please refer to our [paper](https://arxiv.org/abs/2312.08548) or [github](https://github.com/Lavreniuk/EVP) for more details.""" with gr.Blocks() as demo: gr.Markdown(title) gr.Markdown(description) with gr.Tab("Depth Prediction"): create_demo(model, device) gr.HTML('''


You can duplicate this Space to skip the queue:Duplicate Space

visitors

''') demo.queue().launch(share=True) if __name__ == '__main__': main()