Update app.py

app.py

@@ -1,5 +1,7 @@
 import os
+import cv2
 import argparse
+import glob
 import numpy as np
 import os
 import torch
@@ -13,7 +15,6 @@ from models.safmn_arch import SAFMN
 from gradio_imageslider import ImageSlider
 
  
-########################################## Gradio inference  ###################################
 pretrain_model_url = {
 	'safmn_x2': 'https://github.com/sunny2109/SAFMN/releases/download/v0.1.0/SAFMN_L_Real_LSDIR_x2-v2.pth',
 	'safmn_x4': 'https://github.com/sunny2109/SAFMN/releases/download/v0.1.0/SAFMN_L_Real_LSDIR_x4-v2.pth',
@@ -171,7 +172,8 @@ title = "SAFMN for Real-world SR"
 description = ''' ### Spatially-Adaptive Feature Modulation for Efficient Image Super-Resolution - ICCV 2023
 #### Long Sun](https://github.com/sunny2109), [Jiangxin Dong](https://scholar.google.com/citations?user=ruebFVEAAAAJ&hl=zh-CN&oi=ao), [Jinhui Tang](https://scholar.google.com/citations?user=ByBLlEwAAAAJ&hl=zh-CN), and [Jinshan Pan](https://jspan.github.io/)
 #### [IMAG Lab](https://imag-njust.net/), Nanjing University of Science and Technology
-####  Drag the slider on the super-resolution image left and right to see the changes in the image details. SAFMN performs x2/x4 upscaling on the input image.
+#### Drag the slider on the super-resolution image left and right to see the changes in the image details.
+#### SAFMN performs x2/x4 upscaling on the input image.  If the input image is larger than 720P, it is recommended to use Memory-efficient inference.
 <br>
 ### If our work is useful for your research, please consider citing:
 <code>
@@ -214,9 +216,9 @@ demo = gr.Interface(
     fn=inference,
     inputs=[
         gr.Image(value="real_testdata/004.png", type="pil", label="Input"),
-        gr.Number(value=2, label="Upscaling factor (up to 4)"),
-		gr.Checkbox(value=False, label="Memory-efficient inference"),
-        gr.Checkbox(value=False, label="Color correction"),
+        gr.Number(default=10, label="Upscaling factor (up to 4)"),
+		gr.Checkbox(default=False, label="Memory-efficient inference"),
+        gr.Checkbox(default=False, label="Color correction"),
     ],
     outputs=ImageSlider(label="Super-Resolved Image", 
                         type="pil",