working demo

app.py

@@ -0,0 +1,197 @@
+import gradio as gr
+from PIL import Image, ImageFilter
+import numpy as np
+import torch
+from torch.autograd import Variable
+from torchvision import transforms
+import torch.nn.functional as F
+import gdown
+import os
+
+os.system("git clone https://github.com/xuebinqin/DIS")
+os.system("mv DIS/IS-Net/* .")
+
+# project imports
+from data_loader_cache import normalize, im_reader, im_preprocess 
+from models import *
+
+#Helpers
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
+# Download official weights
+if not os.path.exists("saved_models"):
+    os.mkdir("saved_models")
+    MODEL_PATH_URL = "https://drive.google.com/uc?id=1KyMpRjewZdyYfxHPYcd-ZbanIXtin0Sn"
+    gdown.download(MODEL_PATH_URL, "saved_models/isnet.pth", use_cookies=False)
+    
+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)
+        return image
+
+
+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)
+    im, im_shp = im_preprocess(im, hypar["cache_size"])
+    im = torch.divide(im,255.0)
+    shape = torch.from_numpy(np.array(im_shp))
+    return transform(im).unsqueeze(0), shape.unsqueeze(0) # make a batch of image, shape
+
+
+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):
+                layer.float()
+
+    net.to(device)
+
+    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):
+    '''
+    Given an Image, predict the mask
+    '''
+    net.eval()
+
+    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) # 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) # 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 = {} # paramters for inferencing
+
+
+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
+
+##  choose floating point accuracy --
+hypar["model_digit"] = "full" ## indicates "half" or "full" accuracy of float number
+hypar["seed"] = 0
+
+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 infer_mask(image: Image):
+  image_path = image
+  
+  image_tensor, orig_size = load_image(image_path, hypar) 
+  mask = predict(net, image_tensor, orig_size, hypar, device)
+  
+  return Image.fromarray(mask).convert("L")
+
+def blur(image_set: list, blur_amount: int):
+    blurred_image = image_set[0].filter(ImageFilter.GaussianBlur(blur_amount))
+
+    return Image.composite(image_set[0], blurred_image, image_set[1])
+
+
+with gr.Blocks() as interface:
+    default_im = Image.open("newman.jpg").convert("RGB")
+    default_mask = Image.open("newman_mask.jpg").convert("RGB")
+    examples_list = [os.path.join(os.path.dirname(__file__), "newman.jpg"),
+                os.path.join(os.path.dirname(__file__), "abbey.jpg"),
+                os.path.join(os.path.dirname(__file__), "julia.jpg")
+                ]
+
+    current_images = gr.State([default_im, default_mask])
+    mask_toggle = gr.State(False)
+
+    gr.Markdown(
+        """
+        ### Intelligent Photo Blur Using Dichotomous Image Segmentation
+
+        This app leverages the machine learning engine built by Xuebin Qin (https://github.com/xuebinqin/DIS) to mask the prominent subject within a photograph.
+        The mask is used to keep the subject in clear focus while an adjustable slider is available to interactively blur the background.
+        To use, upload a photo and press the run button. You can adjust the level of blur through the slider and view the mask using the "Show Generated Mask" button.
+        """
+    )
+    with gr.Row():
+        with gr.Column():
+            input_image = gr.Image(value=default_im, type='filepath')
+            run_button = gr.Button()
+            gr.Examples(inputs=input_image, examples=examples_list)
+        with gr.Column():
+            output_image = gr.Image()
+            blur_slider = gr.Slider(0, 16, 5, step=1, label="Blur Amount")
+            mask_button = gr.Button(value="Show Generated Mask")
+            mask_image = gr.Image(value=default_mask, visible=False)
+
+        def run(image: Image, current_images: gr.State):
+            im_rgb = Image.open(image).convert("RGB")
+            mask = infer_mask(image)
+
+            return (
+                blur([im_rgb, mask], 5),
+                mask,
+                [im_rgb, mask]
+            )
+        
+        def reset_slider():
+            return gr.update(value=5)
+
+        def show_mask(mask_toggle: gr.State):
+            if mask_toggle == True:
+                return gr.update(visible=False)
+            else:
+                return gr.update(visible=True)
+
+        def toggle_mask(mask_toggle: gr.State):
+            if mask_toggle == True:
+                return False
+            else:
+                return True
+
+        run_button.click(run, [input_image, current_images], [output_image, mask_image, current_images])
+        run_button.click(reset_slider, outputs=blur_slider)
+        blur_slider.change(blur, [current_images, blur_slider], output_image, show_progress=False)
+        mask_button.click(show_mask, inputs=mask_toggle, outputs=mask_image)
+        mask_button.click(toggle_mask, inputs=mask_toggle, outputs=mask_toggle)
+
+interface.launch()

requirements.txt

@@ -0,0 +1,8 @@
+torch
+torchvision
+requests
+gdown
+matplotlib
+opencv-python
+Pillow==8.0.0
+scikit-image==0.15.0