import gradio as gr import numpy as np import spaces import torch import rembg from PIL import Image from functools import partial import logging import os import shlex import subprocess import tempfile import time subprocess.run(shlex.split('pip install wheel/torchmcubes-0.1.0-cp310-cp310-linux_x86_64.whl')) from tsr.system import TSR from tsr.utils import remove_background, resize_foreground, to_gradio_3d_orientation STEP1_HEADER = """ # Step 1: Generate the 3D Mesh For this step, we use TripoSR, an open-source model for **fast** feedforward 3D reconstruction from a single image, developed in collaboration between [Tripo AI](https://www.tripo3d.ai/) and [Stability AI](https://stability.ai/). During this step, you need to upload an image of what you want to generate a 3D Model from. ## 💡 Tips - If there's a background, ✅ Remove background. - If you find the result is unsatisfied, please try to change the foreground ratio. It might improve the results. """ # These part of the code (check_input_image and preprocess were taken from https://huggingface.co/spaces/stabilityai/TripoSR/blob/main/app.py) if torch.cuda.is_available(): device = "cuda:0" else: device = "cpu" model = TSR.from_pretrained( "stabilityai/TripoSR", config_name="config.yaml", weight_name="model.ckpt", ) model.renderer.set_chunk_size(131072) model.to(device) rembg_session = rembg.new_session() def check_input_image(input_image): if input_image is None: raise gr.Error("No image uploaded!") def preprocess(input_image, do_remove_background, foreground_ratio): def fill_background(image): image = np.array(image).astype(np.float32) / 255.0 image = image[:, :, :3] * image[:, :, 3:4] + (1 - image[:, :, 3:4]) * 0.5 image = Image.fromarray((image * 255.0).astype(np.uint8)) return image if do_remove_background: image = input_image.convert("RGB") image = remove_background(image, rembg_session) image = resize_foreground(image, foreground_ratio) image = fill_background(image) else: image = input_image if image.mode == "RGBA": image = fill_background(image) return image @spaces.GPU def generate(image, mc_resolution, formats=["obj", "glb"]): scene_codes = model(image, device=device) mesh = model.extract_mesh(scene_codes, resolution=mc_resolution)[0] mesh = to_gradio_3d_orientation(mesh) mesh_path_glb = tempfile.NamedTemporaryFile(suffix=f".glb", delete=False) mesh.export(mesh_path_glb.name) mesh_path_obj = tempfile.NamedTemporaryFile(suffix=f".obj", delete=False) mesh.apply_scale([-1, 1, 1]) # Otherwise the visualized .obj will be flipped mesh.export(mesh_path_obj.name) return mesh_path_obj.name, mesh_path_glb.name with gr.Blocks() as demo: gr.Markdown(STEP1_HEADER) with gr.Row(variant = "panel"): with gr.Column(): with gr.Row(): input_image = gr.Image( label = "Input Image", image_mode = "RGBA", sources = "upload", type="pil", elem_id="content_image") processed_image = gr.Image(label="Processed Image", interactive=False) with gr.Row(): with gr.Group(): do_remove_background = gr.Checkbox( label="Remove Background", value=True) foreground_ratio = gr.Slider( label="Foreground Ratio", minimum=0.5, maximum=1.0, value=0.85, step=0.05, ) mc_resolution = gr.Slider( label="Marching Cubes Resolution", minimum=32, maximum=320, value=256, step=32 ) with gr.Row(): step1_submit = gr.Button("Generate", elem_id="generate", variant="primary") with gr.Column(): with gr.Tab("OBJ"): output_model_obj = gr.Model3D( label = "Output Model (OBJ Format)", interative = False, ) gr.Markdown("Note: Downloaded object will be flipped in case of .obj export. Export .glb instead or manually flip it before usage.") with gr.Tab("GLB"): output_model_glb = gr.Model3D( label="Output Model (GLB Format)", interactive=False, ) gr.Markdown("Note: The model shown here has a darker appearance. Download to get correct results.") step1_submit.click(fn=check_input_image, inputs=[input_image]).success( fn=preprocess, inputs=[input_image, do_remove_background, foreground_ratio], outputs=[processed_image], ).success( fn=generate, inputs=[processed_image, mc_resolution], outputs=[output_model_obj, output_model_glb], ) demo.queue(max_size=10) demo.launch()