Update app.py

app.py

@@ -2,37 +2,63 @@
 
 # gr.Interface.load("models/Akbartus/Lora360").launch(show_api=True)
 
-import gradio as gr
-import torch
+import tensorflow as tf
+import math
+import numpy as np
 from PIL import Image
-from torchvision.transforms import ToTensor, ToPILImage
-from transformers import AutoModel
+from tensorflow.keras.preprocessing.image import img_to_array
+from huggingface_hub import from_pretrained_keras
+import gradio as gr
 
 # Load the original model
-original_model = gr.Interface.load("models/Akbartus/Lora360")
+original_model = gr.Interface.load("models/Tester/Lora360")
 
 # Load the super-resolution model from Hugging Face's model hub
-super_resolution_model = AutoModel.from_pretrained('keras-io/super-resolution')
+super_resolution_model = from_pretrained_keras("keras-io/super-resolution")
 
 def process_image(input_image):
     # Run the original model
     output_image = original_model(input_image)
 
-    # Transform the output image to tensor
-    output_image = ToTensor()(output_image).unsqueeze(0)
+    # Resize the image to 100x100
+    output_image = output_image.resize((100,100))
+
+    # Convert the image to YCbCr color space
+    ycbcr = output_image.convert("YCbCr")
+    y, cb, cr = ycbcr.split()
+    y = img_to_array(y)
+    y = y.astype("float32") / 255.0
 
-    # Ensure the super-resolution model is in eval mode and perform super-resolution
-    super_resolution_model.eval()
-    with torch.no_grad():
-        super_res_image = super_resolution_model(output_image)
+    # Expand the dimensions of the image for model input
+    input = np.expand_dims(y, axis=0)
+    out = super_resolution_model.predict(input)
 
-    # Convert the output tensor to an image
-    super_res_image = ToPILImage()(super_res_image[0])
-    return super_res_image
+    # Process the output
+    out_img_y = out[0]
+    out_img_y *= 255.0
 
-# Define the Gradio interface
-iface = gr.Interface(fn=process_image, inputs="image", outputs="image")
-iface.launch()
+    # Restore the image in RGB color space.
+    out_img_y = out_img_y.clip(0, 255)
+    out_img_y = out_img_y.reshape((np.shape(out_img_y)[0], np.shape(out_img_y)[1]))
+    out_img_y = Image.fromarray(np.uint8(out_img_y), mode="L")
+    out_img_cb = cb.resize(out_img_y.size, Image.BICUBIC)
+    out_img_cr = cr.resize(out_img_y.size, Image.BICUBIC)
+    out_img = Image.merge("YCbCr", (out_img_y, out_img_cb, out_img_cr)).convert(
+        "RGB"
+    )
 
+    # Return the original resized image and the super-resolved image
+    return (output_image, out_img)
 
+article = "<p style='text-align: center'><a href='https://arxiv.org/abs/1609.05158' target='_blank'>Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network</a></p><center> <a href='https://keras.io/examples/vision/super_resolution_sub_pixel/' target='_blank'>Image Super-Resolution using an Efficient Sub-Pixel CNN</a></p> <center>Contributors: <a href='https://twitter.com/Cr0wley_zz'>Devjyoti Chakraborty</a>|<a href='https://twitter.com/ritwik_raha'>Ritwik Raha</a>|<a href='https://twitter.com/ariG23498'>Aritra Roy Gosthipaty</a></center>"
 
+iface = gr.Interface(
+    fn=process_image,
+    title = " Image Super-resolution",
+    description = "This space is a demo of the keras tutorial 'Image Super-Resolution using an Efficient Sub-Pixel CNN' based on the paper 'Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network' 👀",
+    article = article,
+    inputs=gr.inputs.Image(label="Input Image"),
+    outputs=[gr.outputs.Image(label="Resized 100x100 image"),
+    gr.outputs.Image(label="Super-resolution 300x300 image")
+    ],
+).launch()