Update app.py

app.py

@@ -1,20 +1,24 @@
-import gradio as gr
+import streamlit as st
 import numpy as np
 from PIL import Image
 from transformers import CLIPProcessor, CLIPModel, YolosImageProcessor, YolosForObjectDetection
 import torch
 
-feature_extractor = YolosImageProcessor.from_pretrained("hustvl/yolos-tiny")
-dmodel = YolosForObjectDetection.from_pretrained('hustvl/yolos-tiny')
+st.title("CLIP & CROP")
+st.markdown("**Extract sections of images from your image by using OpenAI's CLIP and Facebooks Detr implemented on HuggingFace Transformers, if the similarity score is not so much, then please consider the prediction to be void.**")
 
-model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
-processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
+with st.spinner("Models are loading"):
+    feature_extractor = YolosImageProcessor.from_pretrained("hustvl/yolos-tiny")
+    dmodel = YolosForObjectDetection.from_pretrained('hustvl/yolos-tiny')
 
-i1 = gr.Image(type="pil", label="Input image")
-i2 = gr.Textbox(label="Description for section to extracted")
-i3 = gr.Number(value=0.96, label="Threshold percentage score")
-o1 = gr.Image(type="pil", label="Extracted Crop part")
-o2 = gr.Textbox(label="Similarity score")
+    model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
+    processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
+ 
+IMAGE_INPUT = st.file_uploader(type="pil", label="Input image")
+TEXT_INPUT = st.text_input(label="Description for section to extracted")
+NUMBER_INPUT = st.number_input(value=0.96, label="Threshold percentage score")
+
+SUBMIT_BUTTON = st.button("SUBMIT")
 
 def extract_image(image, text, prob, num=1):
     
@@ -64,9 +68,8 @@ def extract_image(image, text, prob, num=1):
     fi = sorted(final_ims, key=lambda item: item.get("score"), reverse=True)
     return fi[0]['image'], fi[0]['score']
 
-title = "ClipnCrop"
-description = "<p style= 'color:white'>Extract sections of images from your image by using OpenAI's CLIP and Facebooks Detr implemented on HuggingFace Transformers, if the similarity score is not so much, then please consider the prediction to be void.</p>" 
-examples=[['ex3.jpg', 'black bag', 0.96],['ex2.jpg', 'man in red dress', 0.85]]
-article = "<p style= 'color:white; text-align:center;'><a href='https://github.com/Vishnunkumar/clipcrop' target='_blank'>clipcrop</a></p>"
-gr_app = gr.Interface(fn=extract_image, inputs=[i1, i2, i3], outputs=[o1, o2], title=title, description=description, article=article, examples=examples, concurrency_limit=20)
-gr_app.launch()
+if SUBMIT_BUTTON:
+    imageOutput, scoreOutput = extract(IMAGE_INPUT, TEXT_INPUT, NUMBER_INPUT)
+    st.image(imageOutput, caption="Cropped Image")
+    st.markdown("*Confidence Score:*")
+    st.success(scoreOutput)