Upload Mini-Vision-V3 demo

Mini-Vision-V3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31ef54c30c43db9ab2954812d65cb34801f5583964ebb6a90195246f867a2036
+size 1612075

demo.py

@@ -0,0 +1,46 @@
+import gradio
+import torch
+import torchvision
+from model import MiniVisionV3
+from PIL import Image, ImageOps
+
+
+old_classes = {'0': 0, '1': 1, '2': 2, '3': 3, '4': 4, '5': 5, '6': 6, '7': 7, '8': 8, '9': 9, 'A': 10, 'B': 11, 'C': 12, 'D': 13, 'E': 14, 'F': 15, 'G': 16, 'H': 17, 'I': 18, 'J': 19, 'K': 20, 'L': 21, 'M': 22, 'N': 23, 'O': 24, 'P': 25, 'Q': 26, 'R': 27, 'S': 28, 'T': 29, 'U': 30, 'V': 31, 'W': 32, 'X': 33, 'Y': 34, 'Z': 35, 'a': 36, 'b': 37, 'd': 38, 'e': 39, 'f': 40, 'g': 41, 'h': 42, 'n': 43, 'q': 44, 'r': 45, 't': 46}
+classes = {v: k for k, v in old_classes.items()}
+
+transform = torchvision.transforms.Compose([
+    torchvision.transforms.Resize(28),
+    torchvision.transforms.ToTensor()])
+
+def load_model():
+    minivisionv3 = MiniVisionV3()
+    state_dict = torch.load("Mini-Vision-V3.pth", weights_only=False)
+    minivisionv3.load_state_dict(state_dict)
+    minivisionv3.eval()
+    return minivisionv3
+
+minivisionv3 = load_model()
+
+def inference(img):
+    img_convert = ImageOps.invert(img["composite"])
+    input = transform(img_convert)
+    input = input.unsqueeze(0)
+
+    with torch.no_grad():
+        outputs = minivisionv3(input)
+        prob = torch.softmax(outputs, 1)
+
+    result = {}
+    for i in range(47):
+        result[str(classes[i])] = prob[0][i].item()
+    return result
+
+
+demo = gradio.Interface(fn=inference,
+                        inputs=gradio.Sketchpad(height=560, width=560, image_mode="L", label="Draw Here", type="pil"),
+                        outputs=gradio.Label(label="Results"),
+                        title="Mini-Vision-V3",
+                        description="A lightweight CNN (0.4M params) trained on EMNIST Balanced for handwritten character recognition.")
+
+if __name__ == '__main__':
+    demo.launch()

model.py

@@ -0,0 +1,44 @@
+import torch
+from torch import nn
+
+
+class MiniVisionV3(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.model = nn.Sequential(
+            nn.Conv2d(1, 32, 3, padding=1),
+            nn.BatchNorm2d(32),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+
+            nn.Conv2d(32, 64, 3, padding=1),
+            nn.BatchNorm2d(64),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+
+            nn.Conv2d(64, 128, 3, padding=1),
+            nn.BatchNorm2d(128),
+            nn.ReLU(),
+            nn.MaxPool2d(2),
+
+            nn.Flatten(),
+            nn.Linear(1152, 256),
+            nn.ReLU(),
+            nn.Dropout(0.3),
+            nn.Linear(256, 47),
+        )
+
+    def forward(self, x):
+        x = self.model(x)
+        return x
+
+if __name__ == '__main__':
+    minivisionv3 = MiniVisionV3()
+
+    total_params = sum(param.numel() for param in minivisionv3.parameters())
+    print(f"Total params: {total_params / 1000000: .2f}M")
+
+    # with torch.no_grad():
+    #     input = torch.randn(256, 1, 28, 28)
+    #     output = minivisionv3(input)
+    #     print(output)