How reigns21s062/computer_vision works
Main Components
• Input: Images containing signboards and text.
• Object Detection Module: Uses YOLOv5 to locate and classify signboards (both Hindi and English).
• Recognition Module: Performs OCR using a hybrid CNN-RNN pipeline.
• CNN Component – InceptionV3 for feature extraction.
• RNN Component – GRU for sequential character prediction.
• (Optional) Pre-processing/Post-processing: Implied data cleaning, image normalization, and result aggregation layers (even if only mentioned in notebooks).
Relationships and Data Flow
• InceptionV3 (for extracting visual features)
• GRU (for parsing the sequential features into recognized text).
• Finally, lead to an “Output” component, which represents the recognized text.
Architectural Patterns and Technologies
• That YOLOv5 is handling object detection, a key component that isolates signboards in the images.
• That the recognition pipeline is modular and relies on deep learning components: CNN (InceptionV3) for image feature extraction and RNN (GRU) for sequence modeling.
Overall Structure
• On one branch, the detection pipeline (YOLOv5) extracts and classifies text areas.
• On the second branch, the recognition pipeline applies the CNN-RNN model (with InceptionV3 and GRU) to perform OCR on the cropped areas.
• Finally, both pipelines converge into the final output, which could be a consolidated view of recognized text information and detected signboard classifications.