app.py

import os
import spaces
import glob
import argparse
import shutil
from PIL import Image
import PIL
import gradio as gr
import random
import numpy as np
from diffusion import generate_latent
from vq_vae import create_mask
from huggingface_hub import snapshot_download

from huggingface_hub import login
login(token = os.getenv('HF_TOKEN'))

model_dir = snapshot_download(
    repo_id="srijaydeshpande/diffusion"
)

@spaces.GPU(duration=120)
def create_image(cancer_type):
    tmp_dir = "./tmp"
    if os.path.exists(tmp_dir):
        shutil.rmtree(tmp_dir)
    os.makedirs(tmp_dir)
    generate_latent(model_dir, cancer_type, tmp_dir)
    create_mask(model_dir, "./tmp", "./tmp/test_masks")
    if(cancer_type == 'benign'):
        diffusion_model_name = 'diffusion_dp_benign'
    else:
        diffusion_model_name = 'diffusion_dp'
    os.system('python pix2pixhd_test.py --name ' + diffusion_model_name + ' --dataroot ./tmp --label_nc 0 --results_dir ./tmp --no_instance --resize_or_crop none --checkpoints_dir ' + model_dir)
    image_dir = "./tmp/" + diffusion_model_name + "/test_latest/images"
    input_label_image = Image.open(os.path.join(image_dir, "sample_input_label.jpg"))
    synthesized_image = Image.open(os.path.join(image_dir, "sample_synthesized_image.jpg"))
    return input_label_image, synthesized_image


demo = gr.Interface(
    create_image,
    inputs=gr.Radio(choices=["benign", "malignant"], label="Choose Type", value="benign"),
    outputs=[gr.Image(), gr.Image()],
    title="Diffusion based Image Generation"
)

demo.launch()

# create_image('benign')