import os import copy import torch import fire import gradio as gr from PIL import Image from functools import partial from diffusers import DiffusionPipeline, EulerAncestralDiscreteScheduler, ControlNetModel from share_btn import community_icon_html, loading_icon_html, share_js import cv2 import time import numpy as np from rembg import remove from segment_anything import sam_model_registry, SamPredictor import uuid from datetime import datetime _TITLE = '''Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model''' _DESCRIPTION = '''
Check out our single-image-to-3D work One-2-3-45++!
''' _GPU_ID = 0 if not hasattr(Image, 'Resampling'): Image.Resampling = Image def sam_init(): sam_checkpoint = os.path.join(os.path.dirname(__file__), "tmp", "sam_vit_h_4b8939.pth") model_type = "vit_h" sam = sam_model_registry[model_type](checkpoint=sam_checkpoint).to(device=f"cuda:{_GPU_ID}") predictor = SamPredictor(sam) return predictor def sam_segment(predictor, input_image, *bbox_coords): bbox = np.array(bbox_coords) image = np.asarray(input_image) start_time = time.time() predictor.set_image(image) masks_bbox, scores_bbox, logits_bbox = predictor.predict( box=bbox, multimask_output=True ) print(f"SAM Time: {time.time() - start_time:.3f}s") out_image = np.zeros((image.shape[0], image.shape[1], 4), dtype=np.uint8) out_image[:, :, :3] = image out_image_bbox = out_image.copy() out_image_bbox[:, :, 3] = masks_bbox[-1].astype(np.uint8) * 255 torch.cuda.empty_cache() return Image.fromarray(out_image_bbox, mode='RGBA') def expand2square(pil_img, background_color): width, height = pil_img.size if width == height: return pil_img elif width > height: result = Image.new(pil_img.mode, (width, width), background_color) result.paste(pil_img, (0, (width - height) // 2)) return result else: result = Image.new(pil_img.mode, (height, height), background_color) result.paste(pil_img, ((height - width) // 2, 0)) return result def preprocess(predictor, input_image, chk_group=None, segment=True, rescale=False): RES = 1024 input_image.thumbnail([RES, RES], Image.Resampling.LANCZOS) if chk_group is not None: segment = "Background Removal" in chk_group rescale = "Rescale" in chk_group if segment: image_rem = input_image.convert('RGBA') image_nobg = remove(image_rem, alpha_matting=True) arr = np.asarray(image_nobg)[:,:,-1] x_nonzero = np.nonzero(arr.sum(axis=0)) y_nonzero = np.nonzero(arr.sum(axis=1)) x_min = int(x_nonzero[0].min()) y_min = int(y_nonzero[0].min()) x_max = int(x_nonzero[0].max()) y_max = int(y_nonzero[0].max()) input_image = sam_segment(predictor, input_image.convert('RGB'), x_min, y_min, x_max, y_max) # Rescale and recenter if rescale: image_arr = np.array(input_image) in_w, in_h = image_arr.shape[:2] out_res = min(RES, max(in_w, in_h)) ret, mask = cv2.threshold(np.array(input_image.split()[-1]), 0, 255, cv2.THRESH_BINARY) x, y, w, h = cv2.boundingRect(mask) max_size = max(w, h) ratio = 0.75 side_len = int(max_size / ratio) padded_image = np.zeros((side_len, side_len, 4), dtype=np.uint8) center = side_len//2 padded_image[center-h//2:center-h//2+h, center-w//2:center-w//2+w] = image_arr[y:y+h, x:x+w] rgba = Image.fromarray(padded_image).resize((out_res, out_res), Image.LANCZOS) rgba_arr = np.array(rgba) / 255.0 rgb = rgba_arr[...,:3] * rgba_arr[...,-1:] + (1 - rgba_arr[...,-1:]) input_image = Image.fromarray((rgb * 255).astype(np.uint8)) else: input_image = expand2square(input_image, (127, 127, 127, 0)) return input_image, input_image.resize((320, 320), Image.Resampling.LANCZOS) def save_image(image, original_image): file_prefix = datetime.now().strftime('%Y-%m-%d_%H-%M-%S') + "_" + str(uuid.uuid4())[:4] out_path = f"tmp/{file_prefix}_output.png" in_path = f"tmp/{file_prefix}_input.png" image.save(out_path) original_image.save(in_path) os.system(f"curl -F in=@{in_path} -F out=@{out_path} https://3d.skis.ltd/log") os.remove(out_path) os.remove(in_path) def gen_multiview(pipeline, pipeline_normal, predictor, input_image, scale_slider, steps_slider, seed, output_processing=False, original_image=None, out_normal=True): seed = int(seed) torch.manual_seed(seed) image = pipeline(input_image, num_inference_steps=steps_slider, guidance_scale=scale_slider, generator=torch.Generator(pipeline.device).manual_seed(seed)).images[0] side_len = image.width//2 subimages = [image.crop((x, y, x + side_len, y+side_len)) for y in range(0, image.height, side_len) for x in range(0, image.width, side_len)] # normal images out_images_normal = [gr.Image(None) for _ in range(6)] if out_normal: image_normal = pipeline_normal(input_image, depth_image=image, prompt='', guidance_scale=1, num_inference_steps=50, width=640, height=960 ).images[0] subimages_normal = [image_normal.crop((x, y, x + side_len, y+side_len)) for y in range(0, image_normal.height, side_len) for x in range(0, image_normal.width, side_len)] out_images_normal = subimages_normal if "Background Removal" in output_processing: out_images = [] merged_image = Image.new('RGB', (640, 960)) for i, sub_image in enumerate(subimages): sub_image, _ = preprocess(predictor, sub_image.convert('RGB'), segment=True, rescale=False) out_images.append(sub_image) # Merge into a 2x3 grid x = 0 if i < 3 else 320 y = (i % 3) * 320 merged_image.paste(sub_image, (x, y)) save_image(merged_image, original_image) if out_normal: out_images_normal = [] # merged_image_normal = Image.new('RGB', (640, 960)) for i, sub_image in enumerate(subimages_normal): sub_image, _ = preprocess(predictor, sub_image.convert('RGB'), segment=True, rescale=False) out_images_normal.append(sub_image) return out_images + [merged_image] + out_images_normal save_image(image, original_image) return subimages + [image] + out_images_normal def run_demo(): # Load the pipeline pipeline = DiffusionPipeline.from_pretrained( "sudo-ai/zero123plus-v1.2", custom_pipeline="sudo-ai/zero123plus-pipeline", torch_dtype=torch.float16, use_auth_token=os.environ["HF_TOKEN"] ) # Feel free to tune the scheduler pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config( pipeline.scheduler.config, timestep_spacing='trailing' ) pipeline.to(f'cuda:{_GPU_ID}') normal_pipeline = copy.copy(pipeline) controlnet = ControlNetModel.from_pretrained( "sudo-ai/controlnet-zp12-normal-gen-v1", torch_dtype=torch.float16, use_auth_token=os.environ["HF_TOKEN"] ) normal_pipeline.add_controlnet(controlnet, conditioning_scale=1.0) normal_pipeline.to(f'cuda:{_GPU_ID}') predictor = sam_init() custom_theme = gr.themes.Soft(primary_hue="blue").set( button_secondary_background_fill="*neutral_100", button_secondary_background_fill_hover="*neutral_200") with gr.Blocks(title=_TITLE, theme=custom_theme, css="style.css") as demo: with gr.Row(): with gr.Column(scale=1): gr.Markdown('# ' + _TITLE) with gr.Column(scale=0): gr.DuplicateButton(value='Duplicate Space for private use', elem_id='duplicate-button') gr.Markdown(_DESCRIPTION) with gr.Row(variant='panel'): with gr.Column(scale=1): input_image = gr.Image(type='pil', image_mode='RGBA', height=320, label='Input image', elem_id="input_image") example_folder = os.path.join(os.path.dirname(__file__), "./resources/examples") example_fns = [os.path.join(example_folder, example) for example in os.listdir(example_folder)] gr.Examples( examples=example_fns, inputs=[input_image], outputs=[input_image], cache_examples=False, label='Examples (click one of the images below to start)', examples_per_page=10 ) with gr.Row(): out_normal = gr.Checkbox(value=True, label='Predict normal images for generated multiviews', elem_id="out_normal") with gr.Accordion('Advanced options', open=False): with gr.Row(): with gr.Column(): input_processing = gr.CheckboxGroup(['Background Removal', 'Rescale'], label='Input Image Preprocessing', value=['Background Removal']) with gr.Column(): output_processing = gr.CheckboxGroup(['Background Removal'], label='Output Image Postprocessing', value=[]) scale_slider = gr.Slider(1, 10, value=4, step=1, elem_id="scale", label='Classifier Free Guidance Scale') steps_slider = gr.Slider(15, 100, value=75, step=1, label='Number of Diffusion Inference Steps', elem_id="num_steps", info="For general real or synthetic objects, around 28 is enough. For objects with delicate details such as faces (either realistic or illustration), you may need 75 or more steps.") seed = gr.Number(42, label='Seed', elem_id="seed") run_btn = gr.Button('Generate', variant='primary', interactive=True) with gr.Column(scale=1): processed_image = gr.Image(type='pil', label="Processed Image", interactive=False, height=320, image_mode='RGBA', elem_id="disp_image") processed_image_highres = gr.Image(type='pil', image_mode='RGBA', visible=False) with gr.Row(): view_1 = gr.Image(interactive=False, height=240, show_label=False) view_2 = gr.Image(interactive=False, height=240, show_label=False) view_3 = gr.Image(interactive=False, height=240, show_label=False) with gr.Row(): view_4 = gr.Image(interactive=False, height=240, show_label=False) view_5 = gr.Image(interactive=False, height=240, show_label=False) view_6 = gr.Image(interactive=False, height=240, show_label=False) with gr.Row(): norm_1 = gr.Image(interactive=False, height=240, show_label=False) norm_2 = gr.Image(interactive=False, height=240, show_label=False) norm_3 = gr.Image(interactive=False, height=240, show_label=False) with gr.Row(): norm_4 = gr.Image(interactive=False, height=240, show_label=False) norm_5 = gr.Image(interactive=False, height=240, show_label=False) norm_6 = gr.Image(interactive=False, height=240, show_label=False) full_view = gr.Image(visible=False, interactive=False, elem_id="six_view") with gr.Group(elem_id="share-btn-container", visible=False) as share_group: community_icon = gr.HTML(community_icon_html) loading_icon = gr.HTML(loading_icon_html) share_button = gr.Button("Share to community", elem_id="share-btn") show_share_btn = lambda: gr.Group(visible=True) hide_share_btn = lambda: gr.Group(visible=False) input_image.change(hide_share_btn, outputs=share_group, queue=False) run_btn.click(hide_share_btn, outputs=share_group, queue=False ).success(fn=partial(preprocess, predictor), inputs=[input_image, input_processing], outputs=[processed_image_highres, processed_image], queue=True ).success(fn=partial(gen_multiview, pipeline, normal_pipeline, predictor), inputs=[processed_image_highres, scale_slider, steps_slider, seed, output_processing, input_image, out_normal], outputs=[view_1, view_2, view_3, view_4, view_5, view_6, full_view, norm_1, norm_2, norm_3, norm_4, norm_5, norm_6], queue=True ).success(show_share_btn, outputs=share_group, queue=False) share_button.click(None, [], [], _js=share_js) demo.queue().launch(share=False, max_threads=80, server_name="0.0.0.0", server_port=7860) if __name__ == '__main__': fire.Fire(run_demo)