import os import json import numpy as np import gradio as gr import cv2 from drexel_metadata.gen_metadata import gen_metadata from PIL import Image def create_temp_file_path(prefix, suffix): with tempfile.NamedTemporaryFile(prefix=prefix, suffix=suffix, delete=False) as tmpfile: return tmpfile.name def run_inference(input_img): # input_mg: NumPy array with the shape (width, height, 3) # Save input_mg as a temporary file tmpfile = create_temp_file_path(prefix="input_", suffix, ".png") im = Image.fromarray(input_img) im.save(tmpfile) # Create temp filenames for output images visfname = create_temp_file_path(prefix="vis_", suffix=".png") maskfname = create_temp_file_path(prefix="mask_", suffix=".png") # Run inference result = gen_metadata(tmpfile, device='cpu', maskfname=maskfname, visfname=visfname) json_metadata = json.dumps(result) # Cleanup os.remove(tempfile) return visfname, maskfname, json_metadata def read_app_header_markdown(): with open('app_header.md') as infile: return infile.read() dm_app = gr.Interface( fn=run_inference, # Input shows markdown explaining and app and a single image upload panel inputs=[ gr.Markdown(read_app_header_markdown()), gr.Image() ], # Output consists of a visualization image, a masked image, and JSON metadata outputs=[ gr.Image(label='visualization'), gr.Image(label='mask'), gr.JSON(label="JSON metadata") ], allow_flagging="never" # Do not save user's results or prompt for users to save the results ) dm_app.launch()