本文分享自华为云社区《绘制一切》，作者： 雨落无痕 。
相关链接：
Notebook案例地址：绘制一切
AI Gallery：https://developer.huaweicloud.com/develop/aigallery/home.html
也可通过AI Gallery，搜索【绘制一切】一键体验！
通过一键点击标记选定对象，即可实现移除指定对象、填补指定对象、替换一切场景，涵盖了包括目标移除、目标填充、背景替换等在内的多种典型图像修补应用场景。
它的整体框架如图所示：
Inpaint Anything结合了 SAM、图像修补模型（例如 LaMa）和 AIGC 模型（例如 Stable Diffusion）等视觉基础模型。
将三个模型结合到一起，我们可以做出很多的功能。本文就实现了在图片/视频中移除一切物体、在图片中填充一切物体和在图片中替换一切背景这三种功能，其具体实现步骤如下：
以下为具体通过ModelArts实现Inpaint Anything的流程。
使用方法：
🔹 本案例需使用 Pytorch-1.8 GPU-P100 及以上规格运行
🔹 点击Run in ModelArts，将会进入到ModelArts CodeLab中，这时需要你登录华为云账号，如果没有账号，则需要注册一个，且要进行实名认证，参考《ModelArts准备工作_简易版》 即可完成账号注册和实名认证。 登录之后，等待片刻，即可进入到CodeLab的运行环境
🔹 出现 Out Of Memory ，请检查是否为您的参数配置过高导致，修改参数配置，重启kernel或更换更高规格资源进行规避❗❗❗
下面让我们从零开始，一起来体验Inpaint Anything绘制一切的乐趣吧！
拷贝代码，并安装依赖库
import os import torch import os.path as osp import moxing as mox path = osp.join(os.getcwd(),'Inpaint-Anything') if not os.path.exists(path): mox.file.copy_parallel('obs://modelarts-labs-bj4-v2/case_zoo/Inpaint-Anything', path) if os.path.exists(path): print('Download success') else: raise Exception('Download Failed') else: print("Model Package already exists!")
!python remove_anything.py \ --input_img ./example/remove-anything/dog.jpg \ --coords_type key_in \ --point_coords 200 450 \ --point_labels 1 \ --dilate_kernel_size 15 \ --output_dir ./results \ --sam_model_type "vit_h" \ --sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \ --lama_config ./lama/configs/prediction/default.yaml \ --lama_ckpt ./pretrained_models/big-lama
import cv2 import matplotlib.pyplot as plt def show_original_image(image_path, modify_image_path): image = cv2.imread(image_path) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) with_point_image = cv2.imread(modify_image_path + 'with_points.png') with_point_image = cv2.cvtColor(with_point_image, cv2.COLOR_BGR2RGB) fig = plt.figure(figsize=(20, 10)) ax1 = fig.add_subplot(1, 2, 1) plt.title('Original image', fontsize=16) ax1.axis('off') ax1.imshow(image) ax2 = fig.add_subplot(1, 2, 2) plt.title('With_Point image', fontsize=16) ax2.axis('off') ax2.imshow(with_point_image) plt.show() def show_modify_image(modify_image_path, image_class): fig = plt.figure(figsize=(20, 15)) save_path = modify_image_path index = 1 for i in range(0,3): for image_item in image_class: file_name = image_item + str(i) + '.png' file_path = save_path + file_name image = cv2.imread(file_path) image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) ax = fig.add_subplot(3,3,index) ax.xaxis.set_visible(False) ax.yaxis.set_visible(False) plt.title(image_item + str(i), fontsize=16) ax.imshow(image) index = index + 1 plt.show()
image_path = './example/remove-anything/dog.jpg' modify_image_path = './results/dog/' image_class = ['with_mask_','mask_','inpainted_with_mask_'] show_original_image(image_path, modify_image_path) show_modify_image(modify_image_path,image_class)
!python fill_anything.py \ --input_img ./example/fill-anything/sample1.png \ --coords_type key_in \ --point_coords 750 500 \ --point_labels 1 \ --text_prompt "a teddy bear on a bench" \ --dilate_kernel_size 50 \ --output_dir ./results \ --sam_model_type "vit_h" \ --sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \ --model_path "stable-diffusion-2-inpainting"
image_path = './example/fill-anything/sample1.png' modify_image_path = './results/sample1/' image_class = ['with_mask_','mask_','filled_with_mask_'] show_original_image(image_path, modify_image_path) show_modify_image(modify_image_path,image_class)
!python replace_anything.py \ --input_img ./example/replace-anything/dog1.png \ --coords_type key_in \ --point_coords 750 500 \ --point_labels 1 \ --text_prompt "sit on the swing" \ --output_dir ./results \ --sam_model_type "vit_h" \ --sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \ --model_path "stable-diffusion-2-inpainting"
image_path = './example/replace-anything/dog1.png' modify_image_path = './results/dog1/' image_class = ['with_mask_','mask_','replaced_with_mask_'] show_original_image(image_path, modify_image_path) show_modify_image(modify_image_path,image_class)
!python remove_anything_video.py \ --input_video ./example/video/paragliding/original_video.mp4 \ --coords_type key_in \ --point_coords 652 162 \ --point_labels 1 \ --dilate_kernel_size 15 \ --output_dir ./results \ --sam_model_type "vit_h" \ --sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \ --lama_config lama/configs/prediction/default.yaml \ --lama_ckpt ./pretrained_models/big-lama \ --tracker_ckpt vitb_384_mae_ce_32x4_ep300 \ --vi_ckpt ./pretrained_models/sttn.pth \ --mask_idx 2 \ --fps 25
from ipywidgets import Output, GridspecLayout from IPython import display filepaths = ["./example/video/paragliding/original_video.mp4","./results/w_mask_15.mp4", "./results/removed_w_mask_15.mp4"] grid = GridspecLayout(1,len(filepaths)) for i, filepath in enumerate(filepaths): out = Output() with out: display.display(display.Video(filepath, embed=True,width=250,height=140)) grid[0, i] = out grid
运行完成后，从左到右依次为原始视频，选中（去除）对象的视频，去除后的视频。
为了方便大家使用一键分割案例，当前增加了Gradio可视化部署案例演示。
示例效果如下：
图片去除
图片填充
背景替换
