ai-card-hub / app.py
GotThatData's picture
Update app.py
8ee319e verified
raw
history blame
13.4 kB
from pydrive2.auth import GoogleAuth
from pydrive2.drive import GoogleDrive
import os
import gradio as gr
from datasets import load_dataset, Dataset
import pandas as pd
from PIL import Image
import pytesseract
import cv2
import numpy as np
import tensorflow as tf
from transformers import LayoutLMv2Processor, LayoutLMv2ForSequenceClassification
import torch
from tqdm import tqdm
import logging
import re
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class CardPreprocessor:
def __init__(self):
# Initialize OCR and models
self.processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
self.ocr_threshold = 0.5
def extract_text_regions(self, image):
"""Extract text regions from the image using OCR"""
try:
# Convert PIL Image to cv2 format
img_cv = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
# Preprocess image for better OCR
gray = cv2.cvtColor(img_cv, cv2.COLOR_BGR2GRAY)
blurred = cv2.GaussianBlur(gray, (5, 5), 0)
thresh = cv2.threshold(blurred, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
# Perform OCR
text = pytesseract.image_to_data(thresh, output_type=pytesseract.Output.DICT)
# Extract relevant information
extracted_info = {
'player_name': None,
'team': None,
'year': None,
'card_number': None,
'brand': None,
'stats': []
}
# Process OCR results
for i, word in enumerate(text['text']):
if word.strip():
conf = int(text['conf'][i])
if conf > 50: # Filter low-confidence detections
# Try to identify year
year_match = re.search(r'19[0-9]{2}|20[0-2][0-9]', word)
if year_match:
extracted_info['year'] = year_match.group()
# Try to identify card number
card_num_match = re.search(r'#\d+|\d+/\d+', word)
if card_num_match:
extracted_info['card_number'] = card_num_match.group()
# Look for common card brands
brands = ['topps', 'upper deck', 'panini', 'fleer', 'bowman']
if word.lower() in brands:
extracted_info['brand'] = word.lower()
# Look for statistics (numbers with common sports stats patterns)
stats_match = re.search(r'\d+\s*(?:HR|RBI|AVG|YDS|TD)', word)
if stats_match:
extracted_info['stats'].append(stats_match.group())
return extracted_info
except Exception as e:
logger.error(f"Error in OCR processing: {str(e)}")
return None
def analyze_card_condition(self, image):
"""Analyze the physical condition of the card"""
try:
# Convert PIL Image to cv2 format
img_cv = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
# Edge detection for corner and edge analysis
edges = cv2.Canny(img_cv, 100, 200)
# Analyze corners
corner_regions = {
'top_left': edges[0:50, 0:50],
'top_right': edges[0:50, -50:],
'bottom_left': edges[-50:, 0:50],
'bottom_right': edges[-50:, -50:]
}
corner_scores = {k: np.mean(v) for k, v in corner_regions.items()}
# Analyze centering
height, width = img_cv.shape[:2]
center_x = width // 2
center_y = height // 2
# Calculate centering score
centering_score = self.calculate_centering(img_cv, center_x, center_y)
condition_info = {
'corner_scores': corner_scores,
'centering_score': centering_score,
'overall_condition': self.calculate_overall_condition(corner_scores, centering_score)
}
return condition_info
except Exception as e:
logger.error(f"Error in condition analysis: {str(e)}")
return None
def calculate_centering(self, image, center_x, center_y):
"""Calculate the centering score of the card"""
try:
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(gray, 50, 150)
# Find contours
contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
if contours:
# Find the largest contour (assumed to be the card)
main_contour = max(contours, key=cv2.contourArea)
x, y, w, h = cv2.boundingRect(main_contour)
# Calculate centering scores
x_score = abs(0.5 - (x + w/2) / image.shape[1])
y_score = abs(0.5 - (y + h/2) / image.shape[0])
return 1 - (x_score + y_score) / 2
return None
except Exception as e:
logger.error(f"Error in centering calculation: {str(e)}")
return None
def calculate_overall_condition(self, corner_scores, centering_score):
"""Calculate overall condition score"""
if corner_scores and centering_score:
corner_avg = sum(corner_scores.values()) / len(corner_scores)
return (corner_avg + centering_score) / 2
return None
def detect_orientation(self, image):
"""Detect if the card is portrait or landscape"""
width, height = image.size
return 'portrait' if height > width else 'landscape'
class DatasetManager:
def __init__(self, local_images_dir="downloaded_cards"):
self.local_images_dir = local_images_dir
self.drive = None
self.dataset_name = "GotThatData/sports-cards"
self.preprocessor = CardPreprocessor()
# Create local directory if it doesn't exist
os.makedirs(local_images_dir, exist_ok=True)
def authenticate_drive(self):
"""Authenticate with Google Drive"""
try:
gauth = GoogleAuth()
gauth.LocalWebserverAuth()
self.drive = GoogleDrive(gauth)
return True, "Successfully authenticated with Google Drive"
except Exception as e:
return False, f"Authentication failed: {str(e)}"
def process_image(self, image_path):
"""Process a single image and extract information"""
try:
with Image.open(image_path) as img:
# Extract text information
text_info = self.preprocessor.extract_text_regions(img)
# Analyze card condition
condition_info = self.preprocessor.analyze_card_condition(img)
# Get orientation
orientation = self.preprocessor.detect_orientation(img)
return {
'text_info': text_info,
'condition_info': condition_info,
'orientation': orientation
}
except Exception as e:
logger.error(f"Error processing image {image_path}: {str(e)}")
return None
def generate_filename(self, info):
"""Generate filename based on extracted information"""
year = info['text_info'].get('year', 'unknown_year')
brand = info['text_info'].get('brand', 'unknown_brand')
number = info['text_info'].get('card_number', '').replace('#', '').replace('/', '_')
if not number:
number = 'unknown_number'
return f"sports_card_{year}_{brand}_{number}"
def download_and_rename_files(self, drive_folder_id):
"""Download files from Google Drive and process them"""
if not self.drive:
return False, "Google Drive not authenticated", []
try:
# List files in the folder
query = f"'{drive_folder_id}' in parents and trashed=false"
file_list = self.drive.ListFile({'q': query}).GetList()
if not file_list:
file = self.drive.CreateFile({'id': drive_folder_id})
if file:
file_list = [file]
else:
return False, "No files found with the specified ID", []
processed_files = []
for i, file in enumerate(tqdm(file_list, desc="Processing files")):
if file['mimeType'].startswith('image/'):
temp_path = os.path.join(self.local_images_dir, f"temp_{i}.jpg")
# Download file
file.GetContentFile(temp_path)
# Process image
info = self.process_image(temp_path)
if info:
# Generate filename based on extracted info
base_filename = self.generate_filename(info)
new_filename = f"{base_filename}.jpg"
final_path = os.path.join(self.local_images_dir, new_filename)
# Rename file
os.rename(temp_path, final_path)
processed_files.append({
'file_path': final_path,
'original_name': file['title'],
'new_name': new_filename,
'image': final_path,
'extracted_info': info['text_info'],
'condition': info['condition_info'],
'orientation': info['orientation']
})
else:
os.remove(temp_path)
return True, f"Successfully processed {len(processed_files)} images", processed_files
except Exception as e:
return False, f"Error processing files: {str(e)}", []
def update_huggingface_dataset(self, processed_files):
"""Update the sports-cards dataset with processed images"""
try:
# Create a DataFrame with the file information
df = pd.DataFrame(processed_files)
# Create a Hugging Face Dataset from the new files
new_dataset = Dataset.from_pandas(df)
try:
# Try to load existing dataset
existing_dataset = load_dataset(self.dataset_name)
# Concatenate with existing dataset if it exists
if 'train' in existing_dataset:
new_dataset = concatenate_datasets([existing_dataset['train'], new_dataset])
except Exception:
logger.info("Creating new dataset")
# Push to Hugging Face Hub
new_dataset.push_to_hub(self.dataset_name, split="train")
return True, f"Successfully updated dataset '{self.dataset_name}' with {len(processed_files)} processed images"
except Exception as e:
return False, f"Error updating Hugging Face dataset: {str(e)}"
def process_pipeline(folder_id):
"""Main pipeline to process images and update dataset"""
manager = DatasetManager()
# Step 1: Authenticate
auth_success, auth_message = manager.authenticate_drive()
if not auth_success:
return auth_message
# Step 2: Download and process files
success, message, processed_files = manager.download_and_rename_files(folder_id)
if not success:
return message
# Step 3: Update Hugging Face dataset
success, hf_message = manager.update_huggingface_dataset(processed_files)
# Create detailed report
report = f"{message}\n{hf_message}\n\nDetailed Processing Report:\n"
for file in processed_files:
report += f"\nFile: {file['new_name']}\n"
report += f"Extracted Info: {file['extracted_info']}\n"
report += f"Condition Score: {file['condition']['overall_condition']:.2f}\n"
report += f"Orientation: {file['orientation']}\n"
report += "-" * 50
return report
# Gradio interface
demo = gr.Interface(
fn=process_pipeline,
inputs=[
gr.Textbox(
label="Google Drive File/Folder ID",
placeholder="Enter the ID from your Google Drive URL",
value="151VOxPO91mg0C3ORiioGUd4hogzP1ujm"
)
],
outputs=gr.Textbox(label="Processing Report"),
title="AI-Powered Sports Cards Processor",
description="Upload card images to automatically extract information, analyze condition, and add to dataset"
)
if __name__ == "__main__":
demo.launch()