run/gradio_ootd.py

import gradio as gr
import os
from pathlib import Path
import sys
import torch
from PIL import Image, ImageOps
import numpy as np
from utils_ootd import get_mask_location
from cloths_db import cloths_map, modeL_db
PROJECT_ROOT = Path(__file__).absolute().parents[1].absolute()
sys.path.insert(0, str(PROJECT_ROOT))

from preprocess.openpose.run_openpose import OpenPose
from preprocess.humanparsing.run_parsing import Parsing
from ootd.inference_ootd_hd import OOTDiffusionHD
from ootd.inference_ootd_dc import OOTDiffusionDC
from preprocess.openpose.annotator.openpose.util import draw_bodypose

# Set default dtype to float64
# torch.set_default_dtype(torch.float16)


openpose_model_hd = OpenPose(0)
parsing_model_hd = Parsing(0)
ootd_model_hd = OOTDiffusionHD(0)

openpose_model_dc = OpenPose(0)
parsing_model_dc = Parsing(0)
ootd_model_dc = OOTDiffusionDC(0)


category_dict = ['upperbody', 'lowerbody', 'dress']
category_dict_utils = ['upper_body', 'lower_body', 'dresses']


example_path = os.path.join(os.path.dirname(__file__), 'examples')
garment_path = os.path.join(os.path.dirname(__file__), 'examples','garment')
model_hd = os.path.join(example_path, 'model/model_1.png')
garment_hd = os.path.join(example_path, 'garment/03244_00.jpg')

model_dc = os.path.join(example_path, 'model/model_8.png')
garment_dc = os.path.join(example_path, 'garment/048554_1.jpg')

openpose_model_dc.preprocessor.body_estimation.model.to('cuda')
ootd_model_dc.pipe.to('cuda')
ootd_model_dc.image_encoder.to('cuda')
ootd_model_dc.text_encoder.to('cuda')

def convert_to_image(image_array):
    if isinstance(image_array, np.ndarray):
        # Normalize the data to the range [0, 255]
        image_array = 255 * (image_array - np.min(image_array)) / (np.max(image_array) - np.min(image_array))
        # Convert to uint8
        image_array = image_array.astype(np.uint8)
        return Image.fromarray(image_array)
    else:
        # Convert to NumPy array first if necessary
        image_array = np.array(image_array)
        # Normalize and convert to uint8
        image_array = 255 * (image_array - np.min(image_array)) / (np.max(image_array) - np.min(image_array))
        image_array = image_array.astype(np.uint8)
        return Image.fromarray(image_array)

# import spaces

# @spaces.GPU
def process_hd(vton_img, garm_img, n_samples, n_steps, image_scale, seed):
    model_type = 'hd'
    category = 0 # 0:upperbody; 1:lowerbody; 2:dress

    with torch.no_grad():
        openpose_model_hd.preprocessor.body_estimation.model.to('cuda')
        ootd_model_hd.pipe.to('cuda')
        ootd_model_hd.image_encoder.to('cuda')
        ootd_model_hd.text_encoder.to('cuda')
        
        garm_img = Image.open(garm_img).resize((768, 1024))
        vton_img = Image.open(vton_img).resize((768, 1024))
        keypoints = openpose_model_hd(vton_img.resize((384, 512)))
        model_parse, _ = parsing_model_hd(vton_img.resize((384, 512)))

        mask, mask_gray = get_mask_location(model_type, category_dict_utils[category], model_parse, keypoints)
        mask = mask.resize((768, 1024), Image.NEAREST)
        mask_gray = mask_gray.resize((768, 1024), Image.NEAREST)
        
        masked_vton_img = Image.composite(mask_gray, vton_img, mask)

        images = ootd_model_hd(
            model_type=model_type,
            category=category_dict[category],
            image_garm=garm_img,
            image_vton=masked_vton_img,
            mask=mask,
            image_ori=vton_img,
            num_samples=n_samples,
            num_steps=n_steps,
            image_scale=image_scale,
            seed=seed,
        )

    return images



# @spaces.GPU
def process_dc(vton_img, garm_img, category):
    model_type = 'dc'
    if category == 'Upper-body':
        category = 0
    elif category == 'Lower-body':
        category = 1
    else:
        category =2

    with torch.no_grad():
        # openpose_model_dc.preprocessor.body_estimation.model.to('cuda')
        # ootd_model_dc.pipe.to('cuda')
        # ootd_model_dc.image_encoder.to('cuda')
        # ootd_model_dc.text_encoder.to('cuda')
        
        garm_img = Image.open(garm_img).resize((768, 1024))
        vton_img = Image.open(vton_img).resize((768, 1024))
        keypoints ,candidate , subset = openpose_model_dc(vton_img.resize((384, 512)))

        # print(len(keypoints["pose_keypoints_2d"]))
        # print(keypoints["pose_keypoints_2d"])

        # person_image = np.asarray(vton_img)


        # print(len(person_image))
        

        # person_image = np.asarray(Image.open(vton_img).resize((768, 1024))) 

        # output = draw_bodypose(canvas=person_image,candidate=candidate, subset=subset )
        # output_image = Image.fromarray(output)
        # output_image.save('keypose.png')



        left_point = keypoints["pose_keypoints_2d"][2]
        right_point = keypoints["pose_keypoints_2d"][5]

        neck_point = keypoints["pose_keypoints_2d"][1]
        hip_point = keypoints["pose_keypoints_2d"][8]



        print(f'left shoulder - {left_point}')
        print(f'right shoulder - {right_point}')
 
        # #find disctance using Euclidian distance
        shoulder_width_pixels = round(np.sqrt( np.power((right_point[0]-left_point[0]),2) + np.power((right_point[1]-left_point[1]),2)),2)

        height_pixels  = round(np.sqrt( np.power((neck_point[0]-hip_point[0]),2) + np.power((neck_point[1]-hip_point[1]),2)),2) *2


        # # Assuming an average human height 
        average_height_cm = 172.72 *1.5

        # Conversion factor from pixels to cm
        conversion_factor = average_height_cm / height_pixels

        # Convert shoulder width to real-world units
        shoulder_width_cm = shoulder_width_pixels * conversion_factor

        print(f'Shoulder width (in pixels): {shoulder_width_pixels}')
        print(f'Estimated height (in pixels): {height_pixels}')
        print(f'Conversion factor (pixels to cm): {conversion_factor}')
        print(f'Shoulder width (in cm): {shoulder_width_cm}')
        print(f'Shoulder width (in INCH): {round(shoulder_width_cm/2.54,1)}')

        model_parse, face_mask = parsing_model_dc(vton_img.resize((384, 512)))

        model_parse_image = convert_to_image(model_parse)
        face_mask_image = convert_to_image(face_mask)

        # Save the images
        model_parse_image.save('model_parse_image.png')
        face_mask_image.save('face_mask_image.png')


        mask, mask_gray = get_mask_location(model_type, category_dict_utils[category], model_parse, keypoints)

        # final_mask = convert_to_image(mask)
        # final_mask.save("final_mask.png")

        # final_mask_grat = convert_to_image(mask_gray)
        # final_mask_grat.save("final_mask_grat.png")

        mask = mask.resize((768, 1024), Image.NEAREST)
        mask_gray = mask_gray.resize((768, 1024), Image.NEAREST)
        # Save the resized masks
        mask.save("mask_resized.png")
        mask_gray.save("mask_gray_resized.png")
        
        masked_vton_img = Image.composite(mask_gray, vton_img, mask)
        masked_vton_img.save("masked_vton_img.png")

        images = ootd_model_dc(
            model_type=model_type,
            category=category_dict[category],
            image_garm=garm_img,
            image_vton=masked_vton_img,
            mask=mask,
            image_ori=vton_img,
            num_samples=1,
            num_steps=10,
            image_scale=  2.0,
            seed=-1,
        )

    return images

# is_upper = False


block = gr.Blocks().queue()
with block:
    with gr.Row():
        gr.Markdown("# ")

    with gr.Row():
        gr.Markdown("## Virtual Trial Room")
    # with gr.Row():
    #     gr.Markdown("")
    with gr.Row():
        with gr.Column():
            vton_img_dc = gr.Image(label="Model", sources='upload', type="filepath", height=384, value=model_dc)

             # Hidden component to store is_upper value
            # is_upper = gr.State(value=True)

            # #set is_upper variable to True when user selects examples from gr.examples upper/lower body
            # def check_image_type(image_path):
            #     if image_path:
            #         filename = os.path.basename(image_path)
            #         image_type = modeL_db.get(filename, "no-dress")  # Default to "no-dress" if not found
            #         return image_type == "no-dress"
            #     return False  # Default to True if no image

            # vton_img_dc.change(fn=check_image_type, inputs=vton_img_dc)

            

           

        example = gr.Examples(
            label="Select for Upper/Lower Body",
            inputs=vton_img_dc,
            examples_per_page=7,
            examples=[
                os.path.join(example_path, 'model/model_8.png'),
                os.path.join(example_path, 'model/049447_0.jpg'),
                os.path.join(example_path, 'model/049713_0.jpg'),
                os.path.join(example_path, 'model/051482_0.jpg'),
                os.path.join(example_path, 'model/051918_0.jpg'),
                os.path.join(example_path, 'model/051962_0.jpg'),
                os.path.join(example_path, 'model/049205_0.jpg'),
            ],
            
            )
    
    
        example = gr.Examples(
            label="Select for Full Body Dress",
            inputs=vton_img_dc,
            examples_per_page=7,
            examples=[
                os.path.join(example_path, 'model/model_9.png'),
                os.path.join(example_path, 'model/052767_0.jpg'),
                os.path.join(example_path, 'model/052472_0.jpg'),
                os.path.join(example_path, 'model/053514_0.jpg'),
                os.path.join(example_path, 'model/053228_0.jpg'),
                os.path.join(example_path, 'model/052964_0.jpg'),
                os.path.join(example_path, 'model/053700_0.jpg'),
            ],
           
            )
        with gr.Column():
            garm_img_dc = gr.Image(label="Garment", sources='upload', type="filepath", height=384, value=garment_dc)
            category_dc = gr.Dropdown(label="Garment category (important option!!!)", choices=["Upper-body", "Lower-body", "Dress"], value="Upper-body")

            def update_category(image_path):
                if image_path:
                    filename = os.path.basename(image_path)
                    return cloths_map.get(filename, None)  # Return None if not found
                return None  # Return None if no image

            garm_img_dc.change(fn=update_category, inputs=garm_img_dc, outputs=category_dc)
            example = gr.Examples(
                label="Examples (upper-body)",
                inputs=garm_img_dc,
          
                
                

                examples_per_page=7,
                examples=[
                    os.path.join(garment_path,'01260_00.jpg'),
                    os.path.join(garment_path,'01430_00.jpg'),
                    os.path.join(garment_path,'02783_00.jpg'),
                    os.path.join(garment_path,'03751_00.jpg'),
                    os.path.join(garment_path,'06429_00.jpg'),
                    os.path.join(garment_path,'06802_00.jpg'),
                    os.path.join(garment_path,'07429_00.jpg'),
                    os.path.join(garment_path,'08348_00.jpg'),
                    os.path.join(garment_path,'09933_00.jpg'),
                    os.path.join(garment_path,'11028_00.jpg'),
                    os.path.join(garment_path,'11351_00.jpg'),
                    os.path.join(garment_path,'11791_00.jpg'),
                    os.path.join(garment_path, '048554_1.jpg'),
                    os.path.join(garment_path, '049920_1.jpg'),
                    os.path.join(garment_path, '049965_1.jpg'),
                    os.path.join(garment_path, '049949_1.jpg'),
                    os.path.join(garment_path, '050181_1.jpg'),
                    os.path.join(garment_path, '049805_1.jpg'),
                    os.path.join(garment_path, '050105_1.jpg'),
                    os.path.join(garment_path, 'male_tshirt1.png'),
                ])
            example = gr.Examples(
                label="Examples (lower-body)",
                inputs=garm_img_dc,
                examples_per_page=7,
                examples=[
                    os.path.join(garment_path, '051827_1.jpg'),
                    os.path.join(garment_path, '051946_1.jpg'),
                    os.path.join(garment_path, '051473_1.jpg'),
                    os.path.join(garment_path, '051515_1.jpg'),
                    os.path.join(garment_path, '051517_1.jpg'),
                    os.path.join(garment_path, '051988_1.jpg'),
                    os.path.join(garment_path, '051412_1.jpg'),
                ])

            
                
            example = gr.Examples(
                label="Examples (dress)",
                inputs=garm_img_dc,
                examples_per_page=7,
                examples=[
                    os.path.join(garment_path, '053290_1.jpg'),
                    os.path.join(garment_path, '053744_1.jpg'),
                    os.path.join(garment_path, '053742_1.jpg'),
                    os.path.join(garment_path, '053786_1.jpg'),
                    os.path.join(garment_path, '053790_1.jpg'),
                    os.path.join(garment_path, '053319_1.jpg'),
                    os.path.join(garment_path, '052234_1.jpg'),
                ])
        with gr.Column():
            result_gallery_dc = gr.Gallery(label='Output', show_label=False, elem_id="gallery", preview=True, scale=1)   
    with gr.Column():
        run_button_dc = gr.Button(value="Run")
        # n_samples_dc = gr.Slider(label="Images", minimum=1, maximum=4, value=1, step=1)
        # n_steps_dc = gr.Slider(label="Steps", minimum=20, maximum=40, value=20, step=1)
        # scale_dc = gr.Slider(label="Scale", minimum=1.0, maximum=12.0, value=5.0, step=0.1)
        # image_scale_dc = gr.Slider(label="Guidance scale", minimum=1.0, maximum=5.0, value=2.0, step=0.1)
        # seed_dc = gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, value=-1)
       
    # ips_dc = [vton_img_dc, garm_img_dc, category_dc] 
   

    ips_dc = [vton_img_dc, garm_img_dc ,category_dc] 
    run_button_dc.click(fn=process_dc, inputs=ips_dc, outputs=[result_gallery_dc])


block.launch(server_name="0.0.0.0", server_port=7860 )