🤖 AI Models Documentation

Complete Guide to Internal AI Models for Technical Drawing Analysis

11 Models
8 Detection Models
9 LitServe
50K+ Lines of Code
💡This repository analyzes technical engineering drawings to automatically extract specifications like tolerances, welding symbols, and surface finish requirements.
📖 Main Guide

Complete documentation explaining what each model does, how they work, and their internal architectures.

  • Detailed architecture explanations
  • How each model works step-by-step
  • Real-world examples
  • Deep learning concepts explained
Read Full Guide →
🏗️ Architecture Reference

Visual diagrams and architectural breakdowns for quick reference. See how data flows through each model.

  • ASCII architecture diagrams
  • Data flow visualizations
  • Component breakdowns
  • Speed comparisons
View Architectures →
⚡ Quick Summary

TL;DR version - quick lookup for which model to use and what it does. Perfect for busy developers.

  • All models at a glance
  • Which model to use for what
  • Performance specs
  • Real-world scenarios
Quick Lookup →

🧠All 11 Models Overview

1. canvas_ocr
Extract dimension tolerances and beveling specifications from drawings
Detection + OCR
2. deep_feature_v2
Convert images to 1024-dimensional vectors for similarity matching
Feature Extraction
3. welding_symbol_detection
Detect and classify welding symbols in technical drawings
Detection
4. material_shape_classification
Classify material shapes (cylinder, cube, plate, etc)
Classification
5. part_table_recognition
Extract and convert parts tables to HTML/CSV format
Complex Pipeline
6. surface_roughness_detection
Extract surface roughness symbols and numeric values (Ra 3.2)
Detection + OCR
7. thickness_estimation
Detect thickness specification symbols in drawings
Detection
8. manufactured_size_estimation
Detect rectangular regions containing size markings
Detection
9. geometric_tolerance_detection
Detect 14 tolerance types and extract their specifications
Detection + OCR
10. drawing_or_not_classification
Binary classification - is this image a technical drawing?
Classification
11. dummy_inference
Test server that returns static response for testing
Testing
Example Processing Pipeline:
Image → Detection Model → Find Objects → Classification → Text Recognition → Structure Output

🛠️Key Technologies

YOLOX

Fast object detection used by 8 models for finding symbols and text in drawings

EfficientNet-B2

CNN for feature extraction - converts images to vectors for similarity matching

BERT & Vision Models

Text recognition and advanced OCR for extracting specifications from drawings

LitServe

High-performance ML serving framework used by 9 models with automatic batching

PyTorch

Deep learning framework powering all neural network models

FastAPI

REST API framework for simpler classification models