Add OCR app

app.py

@@ -0,0 +1,45 @@
+import torch
+from src.model import CRNN
+from PIL import Image as PILImage
+import torchvision.transforms as transforms
+import gradio as gr
+
+# Load CRNN model
+model = CRNN(img_height=32, img_width=100, img_channel=1, num_class=37, rnn_hidden=256)
+model.load_state_dict(torch.load("crnn_gpu.pt", map_location="cpu"))
+model.eval()
+
+alphabet = '0123456789abcdefghijklmnopqrstuvwxyz'
+
+def ctc_decode(preds):
+    preds = preds.argmax(2).transpose(1,0).contiguous().view(-1)
+    decoded = []
+    prev_idx = -1
+    for idx in preds:
+        if idx != prev_idx and idx != 0:
+            decoded.append(alphabet[idx-1])
+        prev_idx = idx
+    return ''.join(decoded)
+
+transform = transforms.Compose([
+    transforms.Grayscale(),
+    transforms.Resize((32,100)),
+    transforms.ToTensor(),
+    transforms.Normalize((0.5,), (0.5,))
+])
+
+def ocr(image: PILImage.Image):
+    img_tensor = transform(image).unsqueeze(0)
+    with torch.no_grad():
+        preds = model(img_tensor)
+        text = ctc_decode(preds)
+    return text
+
+iface = gr.Interface(
+    fn=ocr,
+    inputs=gr.Image(type="pil", interactive=True, label="Upload and crop image"),
+    outputs="text",
+    title="CRNN OCR"
+)
+
+iface.launch()

requirements.txt

@@ -0,0 +1,4 @@
+torch
+torchvision
+gradio
+Pillow