app.py

import gradio as gr
import os
import sys
import cv2
import numpy as np
import torch
from generate import process, load_model, NORMAL_MAP_MODEL, OTHER_MAP_MODEL

# Ensure the utils directory is in the path
sys.path.append(os.path.join(os.path.dirname(__file__), 'utils'))

# Force CPU usage
device = torch.device('cpu')
normal_map_model = load_model(NORMAL_MAP_MODEL, device)
other_map_model = load_model(OTHER_MAP_MODEL, device)
models = [normal_map_model, other_map_model]

# Temporary input and output directories
input_dir = "temp_input"
output_dir = "temp_output"

if not os.path.exists(input_dir):
    os.makedirs(input_dir)
if not os.path.exists(output_dir):
    os.makedirs(output_dir)

def generate_maps(input_image, tile_size=512, seamless=False, mirror=False, replicate=False, ishiiruka=False, ishiiruka_texture_encoder=False):
    # Save the uploaded image to the input directory
    input_path = os.path.join(input_dir, "input.png")
    cv2.imwrite(input_path, cv2.cvtColor(input_image, cv2.COLOR_RGB2BGR))

    # Prepare arguments as in generate.py
    args = type('Args', (), {
        'input': input_dir,
        'output': output_dir,
        'tile_size': tile_size,
        'seamless': seamless,
        'mirror': mirror,
        'replicate': replicate,
        'ishiiruka': ishiiruka,
        'ishiiruka_texture_encoder': ishiiruka_texture_encoder,
        'cpu': True  # Force CPU usage
    })()

    # Create a mock parser and set args (simulating command-line args)
    import argparse
    parser = argparse.ArgumentParser()
    parser.parse_args = lambda: args

    # Process the image
    img = cv2.imread(input_path, cv2.IMREAD_COLOR)

    if seamless:
        img = cv2.copyMakeBorder(img, 16, 16, 16, 16, cv2.BORDER_WRAP)
    elif mirror:
        img = cv2.copyMakeBorder(img, 16, 16, 16, 16, cv2.BORDER_REFLECT_101)
    elif replicate:
        img = cv2.copyMakeBorder(img, 16, 16, 16, 16, cv2.BORDER_REPLICATE)

    img_height, img_width = img.shape[:2]
    do_split = img_height > args.tile_size or img_width > args.tile_size

    from utils.imgops import esrgan_launcher_split_merge, crop_seamless
    if do_split:
        rlts = esrgan_launcher_split_merge(img, lambda x, m: process(x, m, device), models, scale_factor=1, tile_size=args.tile_size)
    else:
        rlts = [process(img, model, device) for model in models]

    if seamless or mirror or replicate:
        rlts = [crop_seamless(rlt) for rlt in rlts]

    normal_map = rlts[0]
    roughness = rlts[1][:, :, 1]
    displacement = rlts[1][:, :, 0]

    # Save and load outputs
    base = "output"
    if args.ishiiruka_texture_encoder:
        r = 255 - roughness
        g = normal_map[:, :, 1]
        b = displacement
        a = normal_map[:, :, 2]
        output = cv2.merge((b, g, r, a))
        cv2.imwrite(os.path.join(output_dir, f'{base}.mat.png'), output)
        normal_map_output = cv2.cvtColor(output, cv2.COLOR_BGRA2RGB)
    else:
        normal_name = f'{base}_Normal.png'
        cv2.imwrite(os.path.join(output_dir, normal_name), normal_map)
        normal_map_output = cv2.cvtColor(normal_map, cv2.COLOR_BGR2RGB)

        rough_name = f'{base}_Roughness.png'
        rough_img = 255 - roughness if args.ishiiruka else roughness
        cv2.imwrite(os.path.join(output_dir, rough_name), rough_img)
        roughness_output = cv2.cvtColor(rough_img, cv2.COLOR_GRAY2RGB)

        displ_name = f'{base}_Displacement.png'
        cv2.imwrite(os.path.join(output_dir, displ_name), displacement)
        displacement_output = cv2.cvtColor(displacement, cv2.COLOR_GRAY2RGB)

    # Return outputs
    outputs = [normal_map_output]
    if not args.ishiiruka_texture_encoder:
        outputs.extend([roughness_output, displacement_output])

    return outputs

# Gradio interface
interface = gr.Interface(
    fn=generate_maps,
    inputs=[
        gr.Image(type="numpy", label="Upload Diffuse Texture"),
        gr.Slider(128, 1024, value=512, step=128, label="Tile Size"),
        gr.Checkbox(label="Seamless"),
        gr.Checkbox(label="Mirror"),
        gr.Checkbox(label="Replicate"),
        gr.Checkbox(label="Ishiiruka Format"),
        gr.Checkbox(label="Ishiiruka Texture Encoder")
    ],
    outputs=[
        gr.Image(label="Normal Map"),
        gr.Image(label="Roughness Map"),
        gr.Image(label="Displacement Map")
    ],
    title="Material Map Generator",
    description="Upload a diffuse texture to generate Normal, Roughness, and Displacement maps using AI. Adjust settings as needed. Note: Running on CPU due to lack of GPU support in this environment, which may result in slower processing."
)

# Launch the interface
interface.launch()