from transformers import CLIPSegProcessor, CLIPSegForImageSegmentation
import gradio as gr
import torch
import matplotlib.pyplot as plt
import cv2

processor = CLIPSegProcessor.from_pretrained("CIDAS/clipseg-rd64-refined")
model = CLIPSegForImageSegmentation.from_pretrained("CIDAS/clipseg-rd64-refined")

def process_image(image, prompts):
  inputs = processor(text=prompts, images=[image] * len(prompts), padding="max_length", return_tensors="pt")
  
  # predict
  with torch.no_grad():
    outputs = model(**inputs)
    preds = outputs.logits.unsqueeze(1)
  
  filename = f"mask.png"
  plt.imsave(filename,torch.sigmoid(preds[1][0]))
  
  img2 = cv2.imread(filename)
  gray_image = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

  (thresh, bw_image) = cv2.threshold(gray_image, 100, 255, cv2.THRESH_BINARY)
  
  # fix color format
  cv2.cvtColor(bw_image, cv2.COLOR_BGR2RGB)
  
  return Image.fromarray(bw_image)
  
title = "Interactive demo: zero-shot image segmentation with CLIPSeg"
description = "Demo for using CLIPSeg, a CLIP-based model for zero- and one-shot image segmentation. To use it, simply upload an image and add a text to mask (identify in the image), or use one of the examples below and click 'submit'. Results will show up in a few seconds."
article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2112.10003'>CLIPSeg: Image Segmentation Using Text and Image Prompts</a> | <a href='https://huggingface.co/docs/transformers/main/en/model_doc/clipseg'>HuggingFace docs</a></p>"

examples = [["a glass", "something to fill", "wood", "a jar"]]
   
interface = gr.Interface(fn=process_image, 
                     inputs=[gr.Image(type="pil"), gr.Textbox(label="What do you want to identify (separated by comma)?")],
                     outputs=gr.Image(type="pil"),
                     title=title,
                     description=description,
                     article=article,
                     examples=examples)
                     
interface.launch(debug=True)