Upload 8 files

app.py

@@ -0,0 +1,83 @@
+import gradio as gr 
+from PIL import Image
+import torch
+import torchvision.transforms as transforms
+import numpy as np
+
+from archs.model import UNet
+
+
+device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+#define some auxiliary functions
+pil_to_tensor = transforms.ToTensor()
+    
+# define some parameters based on the run we want to make
+
+model = UNet()
+
+checkpoints = torch.load('./models/chk_6000.pt', map_location=device)
+
+model.load_state_dict(checkpoints['model_state_dict'])
+
+model = model.to(device)
+
+model.eval()
+
+def load_img (filename):
+    img = Image.open(filename).convert("RGB")
+    img_tensor = pil_to_tensor(img)
+    return img_tensor
+
+def process_img(image):
+    img = np.array(image)
+    img = img / 255.
+    img = img.astype(np.float32)
+    y = torch.tensor(img).permute(2,0,1).unsqueeze(0).to(device)
+
+    with torch.no_grad():
+        x_hat = model(y)
+
+    restored_img = x_hat.squeeze().permute(1,2,0).clamp_(0, 1).cpu().detach().numpy()
+    restored_img = np.clip(restored_img, 0. , 1.)
+
+    restored_img = (restored_img * 255.0).round().astype(np.uint8)  # float32 to uint8
+    return Image.fromarray(restored_img) #(image, Image.fromarray(restored_img))
+
+title = "Efficient Low-Light Enhancement ✏️🖼️ 🤗"
+description = ''' ## [Inpainting for Autonomous Driving](https://github.com/cidautai)
+[Javier Abad Hernández](https://github.com/javierabad01)
+Fundación Cidaut
+> **Disclaimer:** please remember this is not a product, thus, you will notice some limitations.
+**This demo expects an image with some degradations.**
+Due to the GPU memory limitations, the app might crash if you feed a high-resolution image (2K, 4K).
+<br>
+'''
+
+examples = [['examples/inputs/1.png'],
+            ['examples/inputs/2.png'], 
+            ['examples/inputs/3.png'], 
+            ["examples/inputs/4.png"], 
+            ["examples/inputs/5.png"]]
+
+css = """
+    .image-frame img, .image-container img {
+        width: auto;
+        height: auto;
+        max-width: none;
+    }
+"""
+
+demo = gr.Interface(
+    fn = process_img,
+    inputs = [
+            gr.Image(type = 'pil', label = 'input')
+    ],
+    outputs = [gr.Image(type='pil', label = 'output')],
+    title = title,
+    description = description,
+    examples = examples,
+    css = css
+)
+
+if __name__ == '__main__':
+    demo.launch()

archs/model.py

@@ -0,0 +1,97 @@
+import torch
+import torch.nn as nn
+
+
+class AttentionBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, kernel_size=3, padding=1):
+        super(AttentionBlock, self).__init__()
+        self.conv1 = nn.Conv2d(in_channels, out_channels,
+                               kernel_size=kernel_size, padding=padding)
+        self.conv2 = nn.Conv2d(out_channels, out_channels,
+                               kernel_size=kernel_size, padding=padding)
+        self.attn = nn.MultiheadAttention(
+            out_channels, num_heads=8, batch_first=True)
+        self.norm = nn.LayerNorm(out_channels)
+        self.activation = nn.ReLU()
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.activation(x)
+        x = self.conv2(x)
+        b, c, h, w = x.size()
+        x = x.view(b, c, h * w).permute(2, 0, 1)  # Reshape and permute
+        attn_output, _ = self.attn(x, x, x)
+        x = attn_output.permute(1, 2, 0).view(
+            b, c, h, w)  # Revert the permute and reshape
+        x = x.view(b, c, -1)  # Flatten the last two dimensions
+        # Reshape for LayerNorm and apply normalization
+        x = self.norm(x.reshape(b, -1, c))
+        x = x.view(b, c, h, w)  # Reshape back to original
+        return x
+
+
+class UNet(nn.Module):
+    def __init__(self):
+        super(UNet, self).__init__()
+
+        self.encoder = nn.Sequential(
+            nn.Conv2d(3, 32, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+            nn.Conv2d(32, 64, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+            nn.Conv2d(64, 128, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+            nn.Conv2d(128, 256, kernel_size=3, padding=1),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+            nn.Conv2d(256, 512, kernel_size=3, padding=1),
+            nn.ReLU(),
+        )
+
+        self.lstm = nn.LSTM(512, 512, batch_first=True)
+        self.attn_block = AttentionBlock(512, 512)
+
+        self.decoder = nn.Sequential(
+            nn.ConvTranspose2d(1024, 256, kernel_size=2, stride=2),
+            nn.ReLU(),
+            nn.ConvTranspose2d(512, 128, kernel_size=2, stride=2),
+            nn.ReLU(),
+            nn.ConvTranspose2d(256, 64, kernel_size=2, stride=2),
+            nn.ReLU(),
+            nn.ConvTranspose2d(128, 32, kernel_size=2, stride=2),
+            nn.ReLU(),
+            nn.ConvTranspose2d(64, 3, kernel_size=1),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, x):
+        skip_connections = []
+
+        for layer in self.encoder:
+            x = layer(x)
+            skip_connections.append(x)
+            if isinstance(layer, nn.MaxPool2d):
+                skip_connections.pop()
+
+        batch_size, channels, height, width = x.size()
+        x = x.view(batch_size, -1, channels)
+        x, _ = self.lstm(x)
+        x = x.unsqueeze(1)
+        x = x.permute(0, 2, 3, 1)
+        x = x.reshape(batch_size, channels, height, width)
+
+        x = self.attn_block(x)
+
+        skip_connections = skip_connections[::-1]
+
+        for i, layer in enumerate(self.decoder):
+            if isinstance(layer, nn.ConvTranspose2d):
+                x = layer(torch.cat((x, skip_connections[i]), dim=1))
+            else:
+                x = layer(x)
+
+        return x
+

models/chk_6000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:169f189b0dc9275b81d4877d38f14f4a0155f72781a43b62d731df5a41481a1e
+size 130055955