Update all

app.py

@@ -4,9 +4,14 @@ from PIL import Image
 import tempfile
 import os
 import yaml
+import base64
+import evaluate
 
 def resize_image(image, base_height):
 
+    if image.size[1] == base_height:
+        return image
+
     # Calculate aspect ratio
     w_percent = base_height / float(image.size[1])
     w_size = int(float(image.size[0]) * float(w_percent))
@@ -14,7 +19,43 @@ def resize_image(image, base_height):
     # Resize the image
     return image.resize((w_size, base_height), Image.Resampling.LANCZOS)
 
-def predict(input_image: Image.Image):
+# Get images and respective transcriptions from the examples directory
+def get_example_data(folder_path="./examples/"):
+    
+    example_data = []
+    
+    # Get list of all files in the folder
+    all_files = os.listdir(folder_path)
+    
+    # Loop through the file list
+    for file_name in all_files:
+        
+        file_path = os.path.join(folder_path, file_name)
+        
+        # Check if the file is an image (.png)
+        if file_name.endswith(".jpg"):
+            
+            # Construct the corresponding .txt filename (same name)
+            corresponding_text_file_name = file_name.replace(".jpg", ".txt")
+            corresponding_text_file_path = os.path.join(folder_path, corresponding_text_file_name)
+            
+            # Initialize to a default value
+            transcription = "Transcription not found."
+            
+            # Try to read the content from the .txt file
+            try:
+                with open(corresponding_text_file_path, "r") as f:
+                    transcription = f.read().strip()
+            except FileNotFoundError:
+                pass  # If the corresponding .txt file is not found, leave the default value
+            
+            example_data.append([file_path, transcription])
+            
+    return example_data
+
+def predict(input_image: Image.Image, ground_truth):
+
+    cer = None
 
     try:
 
@@ -53,15 +94,6 @@ def predict(input_image: Image.Image):
             except subprocess.CalledProcessError as e:
                 print(f"Command failed with error {e.returncode}, output:\n{e.output}")
 
-            # subprocess.run(f"pylaia-htr-decode-ctc --config {temp_config_path} | tee predict.txt", shell=True, check=True)
-            
-            # Alternative to shell=True (ChatGPT suggestion)
-            # from subprocess import Popen, PIPE
-
-            # # Run the first command and capture its output
-            # p1 = Popen(["pylaia-htr-decode-ctc", "--config", temp_config_path], stdout=PIPE)
-            # output = p1.communicate()[0]
-
             # # Write the output to predict.txt
             # with open('predict.txt', 'wb') as f:
             #     f.write(output)
@@ -74,23 +106,132 @@ def predict(input_image: Image.Image):
             else:
                 print('predict.txt does not exist')
 
-        return prediction
+            if ground_truth is not None and ground_truth.strip() != "":
+
+                # Debug: Print lengths before computing metric
+                print("Number of predictions:", len(prediction))
+                print("Number of references:", len(ground_truth))
+
+                # Check if lengths match
+                if len(prediction) != len(ground_truth):
+
+                    print("Mismatch in number of predictions and references.")
+                    print("Predictions:", prediction)
+                    print("References:", ground_truth)
+                    print("\n")
+
+                cer = cer_metric.compute(predictions=[prediction], references=[ground_truth])
+                # cer = f"{cer:.3f}"
+
+            else:
+
+                cer = "Ground truth not provided"
+
+        return prediction, cer
 
     except subprocess.CalledProcessError as e:
         return f"Command failed with error {e.returncode}"
 
-# Gradio interface
-title = "PyLaia HTR"
-description = "Inference using PyLaia models."
-examples = [["examples/example01.jpg"], ["examples/example02.jpg"]]
-
-iface = gr.Interface(
-    fn=predict,
-    inputs=gr.inputs.Image(type="pil"),
-    outputs=gr.outputs.Textbox(),
-    title=title,
-    description=description,
-    examples=examples
-)
-
-iface.launch()
+# Encode images
+with open("assets/header.png", "rb") as img_file:
+    logo_html = base64.b64encode(img_file.read()).decode('utf-8')
+
+with open("assets/teklia_logo.png", "rb") as img_file:
+    footer_html = base64.b64encode(img_file.read()).decode('utf-8')
+
+title = """
+    <h1 style='text-align: center'> Hugging Face x Teklia: PyLaia HTR demo</p>
+"""
+
+description = """
+    [PyLaia](https://github.com/jpuigcerver/PyLaia) is a device agnostic, PyTorch-based, deep learning toolkit \
+    for handwritten document analysis.
+    This model was trained using PyLaia library on Norwegian historical documents ([NorHand Dataset](https://zenodo.org/record/6542056)) \
+    during the [HUGIN-MUNIN project](https://hugin-munin-project.github.io).
+    * HF `model card`: [Teklia/pylaia-huginmunin](https://huggingface.co/Teklia/pylaia-huginmunin) | \
+    [A Comprehensive Comparison of Open-Source Libraries for Handwritten Text Recognition in Norwegian](https://doi.org/10.1007/978-3-031-06555-2_27)
+"""
+
+examples = get_example_data()
+
+# pip install evaluate
+# pip install jiwer
+cer_metric = evaluate.load("cer")
+
+with gr.Blocks(
+    theme=gr.themes.Soft(),
+    title="PyLaia HTR",
+) as demo:
+
+    gr.HTML(
+        f"""
+        <div style='display: flex; justify-content: center; width: 100%;'>
+            <img src='data:image/png;base64,{logo_html}' class='img-fluid' width='350px'>
+        </div>
+        """
+    )
+
+    #174x60
+
+    title = gr.HTML(title)
+    description = gr.Markdown(description)
+
+    with gr.Row():
+
+        with gr.Column(variant="panel"):
+
+            input = gr.components.Image(type="pil", label="Input image:")
+
+            with gr.Row():
+
+                btn_clear = gr.Button(value="Clear")
+                button = gr.Button(value="Submit")
+
+        with gr.Column(variant="panel"):
+
+            output = gr.components.Textbox(label="Generated text:")
+            ground_truth = gr.components.Textbox(value="", placeholder="Provide the ground truth, if available.", label="Ground truth:")
+            cer_output = gr.components.Textbox(label="CER:")
+
+    with gr.Row():
+
+        with gr.Accordion(label="Choose an example from test set:", open=False):
+            
+            gr.Examples(
+                examples=examples,
+                inputs = [input, ground_truth],
+                label=None,
+            )
+
+    with gr.Row():
+
+        gr.HTML(
+            f"""
+            <div style="display: flex; align-items: center; justify-content: center">
+                <a href="https://teklia.com/" target="_blank">
+                    <img src="data:image/png;base64,{footer_html}" style="width: 100px; height: 80px; object-fit: contain; margin-right: 5px; margin-bottom: 5px">
+                </a>
+                <p style="font-size: 13px">
+                    |    <a href="https://huggingface.co/Teklia">Teklia models on Hugging Face</a>
+                </p>
+            </div>
+            """
+        )
+
+    button.click(predict, inputs=[input, ground_truth], outputs=[output, cer_output])
+    btn_clear.click(lambda: [None, "", "", ""], outputs=[input, output, ground_truth, cer_output])
+
+    # Try to force light mode
+    js = """
+        function () {
+            gradioURL = window.location.href
+            if (!gradioURL.endsWith('?__theme=light')) {
+                window.location.replace(gradioURL + '?__theme=light');
+        }
+    }"""
+
+    demo.load(_js=js)
+
+if __name__ == "__main__":
+
+    demo.launch(favicon_path="teklia_icon_grey.png")

examples/0ca7c28a-6d9e-4bc1-9b77-58dfdffd8b1b_0.txt

@@ -0,0 +1 @@
+lide den tort at jeg blev borte for ham. Maaske var det

examples/0ca7c28a-6d9e-4bc1-9b77-58dfdffd8b1b_1.txt

@@ -0,0 +1 @@
+hans hensigt at opdage Deres bolig, maaske blot at

examples/example01.txt

@@ -0,0 +1 @@
+og Valstad kan vi vist

examples/example02.txt

@@ -0,0 +1 @@
+ikke gjøre Regning paa,

requirements.txt

@@ -1,2 +1,4 @@
 git+https://github.com/jpuigcerver/PyLaia/
-nnutils-pytorch
+nnutils-pytorch
+jiwer==3.0.3
+evaluate==0.4.0

test_img_list.txt

@@ -1 +0,0 @@
-