Update app.py

app.py

@@ -1,67 +1,60 @@
+import cv2
+import gradio as gr
 import os
+from PIL import Image
 import numpy as np
 import torch
 from torch.autograd import Variable
 from torchvision import transforms
 import torch.nn.functional as F
-from flask import Flask, request, jsonify, render_template, send_from_directory
-from PIL import Image
+import gdown
+import matplotlib.pyplot as plt
 import warnings
 warnings.filterwarnings("ignore")
 
-# Clone repository and setup
-if not os.path.exists("DIS"):
-    os.system("git clone https://github.com/xuebinqin/DIS")
-    os.system("mv DIS/IS-Net/* .")
+os.system("git clone https://github.com/xuebinqin/DIS")
+os.system("mv DIS/IS-Net/* .")
 
-# Project imports
+# project imports
 from data_loader_cache import normalize, im_reader, im_preprocess 
 from models import *
 
-# Setup device
+#Helpers
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
 # Download official weights
 if not os.path.exists("saved_models"):
-    os.makedirs("saved_models", exist_ok=True)
-    if not os.path.exists("saved_models/isnet.pth"):
-        if os.path.exists("isnet.pth"):
-            os.rename("isnet.pth", "saved_models/isnet.pth")
-
+    os.mkdir("saved_models")
+    os.system("mv isnet.pth saved_models/")
+    
 class GOSNormalize(object):
+    '''
+    Normalize the Image using torch.transforms
+    '''
     def __init__(self, mean=[0.485,0.456,0.406], std=[0.229,0.224,0.225]):
         self.mean = mean
         self.std = std
 
-    def __call__(self, image):
-        image = normalize(image, self.mean, self.std)
+    def __call__(self,image):
+        image = normalize(image,self.mean,self.std)
         return image
 
-transform = transforms.Compose([GOSNormalize([0.5,0.5,0.5],[1.0,1.0,1.0])])
+
+transform =  transforms.Compose([GOSNormalize([0.5,0.5,0.5],[1.0,1.0,1.0])])
 
 def load_image(im_path, hypar):
     im = im_reader(im_path)
-    
-    # Convert to RGB if image has alpha channel
-    if isinstance(im, np.ndarray):
-        if im.ndim == 3 and im.shape[2] == 4:
-            im = Image.fromarray(im).convert('RGB')
-        elif im.ndim == 3:
-            im = Image.fromarray(im)
-        elif im.ndim == 2:
-            im = Image.fromarray(im).convert('RGB')
-    elif hasattr(im, 'mode') and im.mode == 'RGBA':
-        im = im.convert('RGB')
-    
     im, im_shp = im_preprocess(im, hypar["cache_size"])
-    im = torch.divide(im, 255.0)
+    im = torch.divide(im,255.0)
     shape = torch.from_numpy(np.array(im_shp))
-    return transform(im).unsqueeze(0), shape.unsqueeze(0)
+    return transform(im).unsqueeze(0), shape.unsqueeze(0) # make a batch of image, shape
 
-def build_model(hypar, device):
-    net = hypar["model"]
-    
-    if hypar["model_digit"] == "half":
+
+def build_model(hypar,device):
+    net = hypar["model"]#GOSNETINC(3,1)
+
+    # convert to half precision
+    if(hypar["model_digit"]=="half"):
         net.half()
         for layer in net.modules():
             if isinstance(layer, nn.BatchNorm2d):
@@ -69,110 +62,92 @@ def build_model(hypar, device):
 
     net.to(device)
 
-    if hypar["restore_model"] != "":
-        net.load_state_dict(torch.load(os.path.join(hypar["model_path"], hypar["restore_model"]), 
-                             map_location=device))
-    net.eval()
+    if(hypar["restore_model"]!=""):
+        net.load_state_dict(torch.load(hypar["model_path"]+"/"+hypar["restore_model"], map_location=device))
+        net.to(device)
+    net.eval()  
     return net
 
-def predict(net, inputs_val, shapes_val, hypar, device):
+    
+def predict(net,  inputs_val, shapes_val, hypar, device):
+    '''
+    Given an Image, predict the mask
+    '''
     net.eval()
 
-    if hypar["model_digit"] == "full":
+    if(hypar["model_digit"]=="full"):
         inputs_val = inputs_val.type(torch.FloatTensor)
     else:
         inputs_val = inputs_val.type(torch.HalfTensor)
 
-    inputs_val_v = Variable(inputs_val, requires_grad=False).to(device)
-    ds_val = net(inputs_val_v)[0]
-    pred_val = ds_val[0][0,:,:,:]
-    
-    pred_val = torch.squeeze(F.upsample(torch.unsqueeze(pred_val,0),
-                                      (shapes_val[0][0], shapes_val[0][1]), mode='bilinear'))
+  
+    inputs_val_v = Variable(inputs_val, requires_grad=False).to(device) # wrap inputs in Variable
+   
+    ds_val = net(inputs_val_v)[0] # list of 6 results
+
+    pred_val = ds_val[0][0,:,:,:] # B x 1 x H x W    # we want the first one which is the most accurate prediction
+
+    ## recover the prediction spatial size to the orignal image size
+    pred_val = torch.squeeze(F.upsample(torch.unsqueeze(pred_val,0),(shapes_val[0][0],shapes_val[0][1]),mode='bilinear'))
 
     ma = torch.max(pred_val)
     mi = torch.min(pred_val)
-    pred_val = (pred_val-mi)/(ma-mi)
-
-    if device == 'cuda': 
-        torch.cuda.empty_cache()
-    return (pred_val.detach().cpu().numpy()*255).astype(np.uint8)
+    pred_val = (pred_val-mi)/(ma-mi) # max = 1
 
+    if device == 'cuda': torch.cuda.empty_cache()
+    return (pred_val.detach().cpu().numpy()*255).astype(np.uint8) # it is the mask we need
+    
 # Set Parameters
-hypar = {
-    "model_path": "./saved_models",
-    "restore_model": "isnet.pth",
-    "interm_sup": False,
-    "model_digit": "full",
-    "seed": 0,
-    "cache_size": [1024, 1024],
-    "input_size": [1024, 1024],
-    "crop_size": [1024, 1024],
-    "model": ISNetDIS()
-}
-
-# Build Model
-net = build_model(hypar, device)
+hypar = {} # paramters for inferencing
 
-# Flask app
-app = Flask(__name__)
-app.config['UPLOAD_FOLDER'] = 'uploads'
-app.config['RESULT_FOLDER'] = 'results'
 
-os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
-os.makedirs(app.config['RESULT_FOLDER'], exist_ok=True)
+hypar["model_path"] ="./saved_models" ## load trained weights from this path
+hypar["restore_model"] = "isnet.pth" ## name of the to-be-loaded weights
+hypar["interm_sup"] = False ## indicate if activate intermediate feature supervision
 
-@app.route('/')
-def index():
-    return render_template('index.html')
+##  choose floating point accuracy --
+hypar["model_digit"] = "full" ## indicates "half" or "full" accuracy of float number
+hypar["seed"] = 0
 
-@app.route('/api/process', methods=['POST'])
-def process_image():
-    if 'image' not in request.files:
-        return jsonify({'error': 'No image provided'}), 400
-    
-    file = request.files['image']
-    if file.filename == '':
-        return jsonify({'error': 'No image selected'}), 400
-    
-    # Save uploaded file
-    upload_path = os.path.join(app.config['UPLOAD_FOLDER'], file.filename)
-    file.save(upload_path)
-    
-    try:
-        # Process image (same as original inference function)
-        image_tensor, orig_size = load_image(upload_path, hypar) 
-        mask = predict(net, image_tensor, orig_size, hypar, device)
-        
-        # Create results
-        pil_mask = Image.fromarray(mask).convert('L')
-        im_rgb = Image.open(upload_path).convert("RGB")
-        im_rgba = im_rgb.copy()
-        im_rgba.putalpha(pil_mask)
-        
-        # Save results
-        result_rgba_path = os.path.join(app.config['RESULT_FOLDER'], f"rgba_{file.filename}")
-        result_mask_path = os.path.join(app.config['RESULT_FOLDER'], f"mask_{file.filename}")
-        
-        im_rgba.save(result_rgba_path, format="PNG")
-        pil_mask.save(result_mask_path, format="PNG")
-        
-        return jsonify({
-            'original': f"/uploads/{file.filename}",
-            'rgba': f"/results/rgba_{file.filename}",
-            'mask': f"/results/mask_{file.filename}",
-            'filename': file.filename
-        })
-    except Exception as e:
-        return jsonify({'error': str(e)}), 500
-
-@app.route('/uploads/<filename>')
-def serve_upload(filename):
-    return send_from_directory(app.config['UPLOAD_FOLDER'], filename)
-
-@app.route('/results/<filename>')
-def serve_result(filename):
-    return send_from_directory(app.config['RESULT_FOLDER'], filename)
-
-if __name__ == '__main__':
-    app.run(host='0.0.0.0', port=7860, debug=True)
+hypar["cache_size"] = [1024, 1024] ## cached input spatial resolution, can be configured into different size
+
+## data augmentation parameters ---
+hypar["input_size"] = [1024, 1024] ## mdoel input spatial size, usually use the same value hypar["cache_size"], which means we don't further resize the images
+hypar["crop_size"] = [1024, 1024] ## random crop size from the input, it is usually set as smaller than hypar["cache_size"], e.g., [920,920] for data augmentation
+
+hypar["model"] = ISNetDIS()
+
+ # Build Model
+net = build_model(hypar, device)
+
+
+def inference(image):
+  image_path = image
+  
+  image_tensor, orig_size = load_image(image_path, hypar) 
+  mask = predict(net, image_tensor, orig_size, hypar, device)
+  
+  pil_mask = Image.fromarray(mask).convert('L')
+  im_rgb = Image.open(image).convert("RGB")
+  
+  im_rgba = im_rgb.copy()
+  im_rgba.putalpha(pil_mask)
+
+  return [im_rgba, pil_mask]
+
+
+title = "Highly Accurate Dichotomous Image Segmentation"
+description = "This is an unofficial demo for DIS, a model that can remove the background from a given image. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below.<br>GitHub: https://github.com/xuebinqin/DIS<br>Telegram bot: https://t.me/restoration_photo_bot<br>[![](https://img.shields.io/twitter/follow/DoEvent?label=@DoEvent&style=social)](https://twitter.com/DoEvent)"
+article = "<div><center><img src='https://visitor-badge.glitch.me/badge?page_id=max_skobeev_dis_cmp_public' alt='visitor badge'></center></div>"
+
+interface = gr.Interface(
+    fn=inference,
+    inputs=gr.Image(type='filepath'),
+    outputs=[gr.Image(type='filepath', format="png"), gr.Image(type='filepath', format="png")],
+    examples=[['robot.png'], ['ship.png']],
+    title=title,
+    description=description,
+    article=article,
+    flagging_mode="never",
+    cache_mode="lazy",
+    ).queue(api_open=True).launch(show_error=True, show_api=True)