Introduction

Vintage car pricing is subjective: condition, rarity, provenance matter. Professional appraisers use experience; machine learning uses image features. Computer vision extracts car condition, identify rare models, compare to comparables for objective pricing.

Car Condition Assessment from Images

Exterior condition: rust, dents, paint quality, glass clarity. Interior condition: upholstery wear, dashboard cracks, mechanical visibility. Photograph car from multiple angles, extract features via CNN, predict condition score.

Accuracy: 80-85% agreement with professional appraisers. Errors: extreme conditions (pristine restoration vs complete neglect) easier to classify; mid-range conditions harder.

Model Identification and Rarity Assessment

Fine-tuned ResNet for car model/year identification from images. Extract make, model, year. Cross-reference with production volume data: low-production models (rarity) command premium. 1950s Ferrari: 100 cars made = very rare; 1970s Datsun: 100,000 cars made = common.

Feature Extraction for Condition Scoring

Key features affecting vintage car value:

• Paint condition: original (premium), respray (discount), touch-ups (discount)
• Chrome/trim: polished (good), dull (poor), missing (bad)
• Interior: original upholstery (premium), restored (good), mismatched (discount)
• Mechanical: visible cleanliness, original parts vs replacements
• Originality: original engine, transmission, etc. (premium)

Comparable Analysis

Extract features from target car. Search database of sold vintage cars for similar cars. Calculate price per feature (hedonic regression): each feature has price coefficient.

Example: 1965 Mustang, good condition, original interior, respray adds up: base_price + restoration_discount + rarity_premium - respray_discount = predicted_price.

Case Study: Classic Car Market

Price 500 vintage cars sold at auction using image-based features:

• Predicted prices within 10-15% of actual sale price (R² = 0.78)
• Error largest for rare models (limited comparables) and extreme conditions
• Appraisers using images + hedonic model outperformed intuition-only valuation

Rarity Premium Quantification

Vintage car value = base model price + condition adjustments + rarity premium. Rarity measured by production volume: log(production) has negative coefficient (rarer = more valuable). A model with 50 cars made is worth 3-5x more than model with 5000 cars made.

Restoration Quality Assessment

Restored cars worth premium if restoration is high quality, or discount if poor quality. Assess restoration quality: paint quality, panel gaps, interior finish, mechanical alignment. High-quality restorations (professional shops) worth 80-90% of original; poor restorations worth 40-50% of original.

Investment Opportunities

Identify underpriced cars: predicted value > asking price. Cars in good condition, moderate rarity, priced below hedonic prediction are buys. Monitor prices: cars appreciating at 5-10% annually are sound investments.

Limitations and Considerations

Provenance (car history, celebrity ownership, racing history) affects value but not visible in images. Hedge against bias by including metadata: year, mileage, service history. Market preferences shift: muscle cars were unpopular in 1990s, valuable today. Models trained on old data may misvalue current market.