🚗 Used Car Classification Project
📊 Phase 1: Data Preparation
Dataset Processing & Sampling Strategy
Raw Dataset
33,137 images
396 car models
→
Data Cleaning
& Quality Control
↓
Resource
Check
& Planning
↓
Random Sampling
10,000 images
≥20 per class
🚗 Used Car Classification Project
🎯 Phase 2: Initial Approach
YOLO Strategy → Switched to Classification
YOLO Object Detection
Strategy
↓
Roboflow Auto-labeling
Requires Paid Version
↓
Manual Labeling
33K images
Too Large Scale
↓
Switched to
Image Classification
No Labels Required
🚗 Used Car Classification Project
🤖 Phase 3: Classification Architecture
Parallel Model Development

🚀 EfficientNet Pipeline

EfficientNetB7
↓
4+ hrs/epoch
Session Timeouts
↓
EfficientNetB0
Epochs: 50→20
↓
Final
Model A

🏗️ ResNet Pipeline

ResNet50
↓
3+ hrs/epoch
GPU Instability
↓
ResNet50
Epochs: 25→5
↓
Final
Model B
🚗 Used Car Classification Project
⚠️ Phase 4: Technical Constraints
Challenges & Solutions
Colab Resource
Limitations
GPU Session
Instability
Memory
Bottlenecks
Training Duration
Constraints
↓
Checkpoint
System
Lightweight
Models
Epoch
Optimization
Data
Sampling
🚗 Used Car Classification Project
📈 Phase 5: Evaluation & Deployment
Model Performance & Results
Final
Model A
Final
Model B
↓
Test
Predictions
↓
Missing
Predictions
→
Fill with zeros
submission.csv
📊 Dataset: 10K/33K samples
🎯 Classes: 396 vehicle models
⚡ Platform: Colab-optimized
🔧 Method: Transfer Learning
🚗 Used Car Classification Project
🚀 Phase 6: Future Enhancements
Roadmap for Improved Performance
Cloud Setup
(AWS/GCP)
Use Full Dataset
Training
Advanced Models
(ViT, ConvNeXt)
Ensemble
Techniques
Combine Models
(Detection + Classification)