Update model card with representative captcha samples

README.md

@@ -1,55 +1,55 @@
 ---
-library_name: transformers
-pipeline_tag: image-to-text
-base_model: microsoft/trocr-base-printed
-tags:
-- trocr
-- vision-encoder-decoder
-- image-to-text
-- ocr
-- captcha
-- math-captcha
-- synthetic-data
----
-
-# OCR Match Captcha
-
-A TrOCR model fine-tuned to recognize short mathematical captcha
-expressions such as `26+7=?` and `45-6=?`.
+    library_name: transformers
+    pipeline_tag: image-to-text
+    base_model: microsoft/trocr-base-printed
+    tags:
+    - trocr
+    - vision-encoder-decoder
+    - image-to-text
+    - ocr
+    - captcha
+    - math-captcha
+    - synthetic-data
+    ---
+
+    # OCR Match Captcha
+
+    A TrOCR model fine-tuned to recognize short mathematical captcha
+    expressions such as `26+7=?` and `45-6=?`.
 
 <table border="0" cellspacing="24" cellpadding="12">
   <tbody>
     <tr>
-      <td style="border: none; padding: 18px;">
+      <td>
         <img
           src="assets/captcha-handwritten-7-plus-12.png"
-          alt="Handwritten mathematical captcha"
+          alt="Handwritten captcha"
           width="290"
           height="90"
         >
       </td>
-      <td style="border: none; padding: 18px;">
+      <td>
         <img
           src="assets/captcha-clean-10-plus-3.png"
-          alt="Clean mathematical captcha"
+          alt="Clean captcha"
           width="290"
           height="90"
         >
       </td>
     </tr>
     <tr>
-      <td style="border: none; padding: 18px;">
+      <td>
         <img
           src="assets/captcha-distorted-18-plus-5.png"
-          alt="Distorted mathematical captcha"
+          alt="Distorted captcha"
           width="290"
           height="90"
         >
       </td>
-      <td style="border: none; padding: 18px;">
+      <td>
         <img
           src="assets/captcha-omex-37-plus-13.png"
-          alt="OMEX mathematical captcha"
+          alt="OMEX captcha"
           width="290"
           height="90"
         >
@@ -58,161 +58,63 @@ expressions such as `26+7=?` and `45-6=?`.
   </tbody>
 </table>
 
-## Model details
-
-- **Architecture:** Vision Encoder-Decoder / TrOCR
-- **Base model:** `microsoft/trocr-base-printed`
-- **Task:** Image-to-text OCR
-- **Target geometry:** 130x30 RGB images
-- **Model-card image size:** 290x90 pixels
-- **Output:** Mathematical expressions without spaces
-- **Supported operators:** `+` and `-`
-- **Expected format:** `number operator number = ?`
-
-## Training data
-
-The model was fine-tuned on 10,000 target-matched synthetic
-mathematical captcha images.
-
-| Split | Samples |
-|---|---:|
-| Training | 8,500 |
-| Validation | 1,000 |
-| Test | 500 |
-
-The synthetic generator balances `+` and `-`, includes `=` and `?`,
-and simulates:
-
-- Thin and low-contrast arithmetic symbols
-- Pastel circle and ring overlap
-- Background noise
-- Blur and interpolation artifacts
-- JPEG compression
-- Small character displacement
-- Variable character spacing
-- Different foreground contrast levels
-
-Labels contain no spaces and use ASCII `-` for subtraction.
-Visual spacing is applied only while rendering the images.
-
-## Training procedure
-
-The complete Vision Encoder-Decoder model was fine-tuned using an
-operator-weighted sequence loss.
-
-Tokens belonging to `+`, `-`, `=`, and `?` received additional
-weight so digit tokens did not dominate the training objective.
-
-The best checkpoint was selected using validation exact-expression
-accuracy. The held-out test split was evaluated only after the best
-validation checkpoint had been restored.
-
-## Training configuration
-
-| Parameter | Value |
-|---|---:|
-| Epochs | 5 |
-| Learning rate | 5e-6 |
-| Batch size | 8 |
-| Gradient accumulation | 2 |
-| Operator-token weight | 3.0 |
-| Generation beams | 4 |
-| Maximum output length | 32 |
-
-## Synthetic evaluation
-
-| Metric | Validation | Test |
-|---|---:|---:|
-| Exact-expression accuracy | 100.000% | 100.000% |
-| Arithmetic-operator accuracy | 100.000% | 100.000% |
-
-These results measure the synthetic rendering family used during
-fine-tuning. They do not guarantee identical performance on every
-real captcha provider, unseen font, or unseen visual style.
-
-## Usage
-
-```python
-import torch
-from PIL import Image
-from transformers import (
-    TrOCRProcessor,
-    VisionEncoderDecoderModel,
-)
-
-repo_id = "arkhabbazan/ocr-match-captcha"
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-processor = TrOCRProcessor.from_pretrained(
-    repo_id,
-    token=True,
-)
-
-model = VisionEncoderDecoderModel.from_pretrained(
-    repo_id,
-    token=True,
-).to(device).eval()
-
-image = Image.open("captcha.jpeg").convert("RGB")
-
-# Keep the original image aspect ratio.
-pixel_values = processor(
-    images=image,
-    return_tensors="pt",
-).pixel_values.to(device)
-
-with torch.inference_mode():
-    generated_ids = model.generate(
-        pixel_values,
-        num_beams=4,
-        max_length=32,
-    )
-
-prediction = processor.batch_decode(
-    generated_ids,
-    skip_special_tokens=True,
-)[0]
-
-print(prediction.replace(" ", ""))
-```
-
-Example output:
-
-```text
-26+7=?
-```
-
-## Input preprocessing
-
-Pass the original RGB captcha directly to `TrOCRProcessor`.
-
-Do not manually stretch a wide captcha to 130x130. The processor
-performs the model's required resizing and normalization.
-
-The `290x90` size used by the four model-card images is only for
-consistent presentation. It is not a required inference input size.
-
-## Intended use
-
-This model is intended for:
-
-- Short mathematical captcha OCR
-- Authorized captcha-processing workflows
-- OCR experimentation
-- Synthetic-to-real domain adaptation research
-- Arithmetic-symbol recognition evaluation
-
-## Limitations
-
-- Training data is primarily synthetic.
-- Unseen fonts and layouts may reduce accuracy.
-- Very faint or heavily occluded operators may be missed.
-- Visual noise may be confused with `+`, `-`, `=`, or `?`.
-- The model is intended for short expressions, not document OCR.
-- Synthetic evaluation may overestimate real-world performance.
-
-## Responsible use
-
-Use this model only on systems and captcha images that you are
-authorized to process. Users are responsible for complying with
-applicable laws, service terms, and access-control policies.
+    ## Model details
+
+    - **Architecture:** Vision Encoder-Decoder / TrOCR
+    - **Base model:** `microsoft/trocr-base-printed`
+    - **Task:** Image-to-text OCR
+    - **Target geometry:** 130x30 RGB images
+    - **Output:** Mathematical expressions without spaces
+
+    ## Training data
+
+    The model was fine-tuned on 10,000 target-matched synthetic images:
+
+    | Split | Samples |
+    |---|---:|
+    | Training | 8,500 |
+    | Validation | 1,000 |
+    | Test | 500 |
+
+    The generator balances `+` and `-`, includes `=` and `?`, and simulates
+    thin low-contrast symbols, pastel circle overlap, blur, JPEG compression,
+    background noise, and small character displacement.
+
+    ## Training configuration
+
+    | Parameter | Value |
+    |---|---:|
+    | Epochs | 5 |
+    | Learning rate | 5e-6 |
+    | Batch size | 8 |
+    | Gradient accumulation | 2 |
+    | Operator-token weight | 3.0 |
+    | Generation beams | 4 |
+    | Maximum output length | 32 |
+
+    ## Usage
+
+    ```python
+    from PIL import Image
+    from transformers import TrOCRProcessor, VisionEncoderDecoderModel
+
+    repo_id = "arkhabbazan/ocr-match-captcha"
+    processor = TrOCRProcessor.from_pretrained(repo_id, token=True)
+    model = VisionEncoderDecoderModel.from_pretrained(repo_id, token=True)
+
+    image = Image.open("captcha.jpeg").convert("RGB")
+    pixel_values = processor(images=image, return_tensors="pt").pixel_values
+    generated_ids = model.generate(pixel_values, num_beams=4, max_length=32)
+    prediction = processor.batch_decode(
+        generated_ids, skip_special_tokens=True
+    )[0]
+    print(prediction.replace(" ", ""))
+    ```
+
+    ## Limitations
+
+    - Training data is primarily synthetic.
+    - Unseen fonts and layouts may reduce accuracy.
+    - Very faint or heavily occluded operators may be missed.
+    - The model is intended for short expressions, not document OCR.
+    - Usage must be authorized and compliant with applicable policies.