Softy-lines's picture
Update README.md
da79933 verified
---
license: apache-2.0
language:
- en
metrics:
- accuracy
library_name: adapter-transformers
pipeline_tag: object-detection
---
# Model Card for Pixelated Captcha Digit Detection
## Model Details
- **License:** Apache-2.0
- **Developed by:** Saidi Souhaieb
- **Finetuned from model:** YOLOv8
## Uses
This model is designed to detect pixelated captcha digits by showing bounding boxes and extracting the coordinates of the detections.
## How to Get Started with the Model
```python
from PIL import Image, ImageDraw
from YOLO import YOLO # Assuming YOLO is a class in a module named YOLO
# Load the model
model = YOLO("models/number_finder.pt")
# Get model results
model_results = model(image_path)
img = Image.open(image_path)
draw = ImageDraw.Draw(img)
# Sort results based on the x-coordinate
detected_objects = []
captcha_result = ""
boxes_data = []
for result in model_results:
for box in result.boxes:
box_data = {"xyxy": box.xyxy[0].tolist(),
"label": str(box.cls.item())}
boxes_data.append(box_data)
sorted_boxes_data = sorted(boxes_data, key=lambda x: x['xyxy'][0])
# Remove similar boxes
unique_sorted_boxes_data = [sorted_boxes_data[0]]
for i in range(1, len(sorted_boxes_data)):
current_box = sorted_boxes_data[i]
prev_box = unique_sorted_boxes_data[-1]
# Check if the current box's x-coordinate is significantly different from the previous box
if abs(current_box['xyxy'][0] - prev_box['xyxy'][0]) > 1: # Adjust threshold as needed
unique_sorted_boxes_data.append(current_box)
for box in unique_sorted_boxes_data:
label_name = box["label"]
box_coords = box["xyxy"]
detected_objects.append((box_coords))
draw.rectangle(box_coords, outline="red")
draw.text((box_coords[0], box_coords[1] - 10), label_name, fill="red")
captcha_result += str(int(label_name[0]) - 2)
img.show() # Display the image with bounding boxes and labels
```
## Training Details
### Training Data
Raw Pixel Digit Captcha Data [https://huggingface.co/datasets/Softy-lines/Raw-Pixel-Digit-Captcha]
## Model Card Authors
[Saidi Souhaieb]