Skip to content

Latest commit

 

History

History
125 lines (93 loc) · 3.11 KB

README.md

File metadata and controls

125 lines (93 loc) · 3.11 KB

EfficientNet for Terrain Classification

Overview

This project implements EfficientNet-B0 for terrain classification, achieving high accuracy in distinguishing between different terrain types.

Model Performance

  • Validation Accuracy: 91.93%
  • Test Accuracy: 92.40%

Model Architecture

EfficientNet is a convolutional neural network architecture that uses compound scaling to balance network depth, width, and resolution for optimal performance.

Key Features

  1. Compound Scaling: Uniformly scales depth/width/resolution using a compound coefficient.
  2. Mobile Inverted Bottleneck Convolution (MBConv): Efficient building block that reduces parameters while maintaining performance.
  3. Squeeze-and-Excitation (SE) blocks: Improves channel interdependencies at minimal computational cost.

Architectural Diagram

Input Image
    │
    ▼
Stem Conv3x3
    │
    ▼
MBConv1, 3x3
    │
    ▼
MBConv6, 3x3
    │
    ▼
MBConv6, 5x5
    │
    ▼
MBConv6, 3x3
    │
    ▼
MBConv6, 5x5
    │
    ▼
MBConv6, 5x5
    │
    ▼
MBConv6, 3x3
    │
    ▼
Conv1x1 & Pooling
    │
    ▼
Fully Connected
    │
    ▼
Output (5 classes)

Each MBConv block includes Squeeze-and-Excitation optimization.

Model Specifications (EfficientNet-B0)

Metric Value
Parameters ~4 million
Model Size ~15.6 MB
Memory Usage ~34.7 MB
CPU Inference Speed ~30-60ms per image
GPU Inference Speed ~10-15ms per image

Note: Actual speed may vary based on hardware and image size.

Implementation Details

Training Process

  1. Input Processing: Images are resized to 224x224 and normalized.
  2. Feature Extraction: EfficientNet backbone extracts hierarchical features.
  3. Global Pooling: Features are pooled to create a fixed-size representation.
  4. Classification: Fully connected layer maps features to 5 terrain classes.
  5. Output: Softmax activation provides class probabilities.

Key Training Parameters

  • Optimizer: Adam
  • Learning Rate: 1e-4
  • Batch Size: 32
  • Loss Function: Cross-entropy loss
  • Number of Epochs: 25
  • Early Stopping: Based on validation accuracy
  • Data Augmentation: Horizontal flip, random crop, rotation

Training and Optimization

We use transfer learning with a pre-trained EfficientNet-B0 model:

  1. Transfer Learning: Pre-trained weights are loaded and initially frozen.
  2. Fine-tuning: Top layers are unfrozen after the first few epochs.
  3. Batch Normalization: Layers remain in training mode during fine-tuning.

Evaluation Metrics

  • Accuracy
  • Confusion Matrix
  • Precision, Recall, F1 Score
  • Loss

Usage

To use the model for terrain classification:

  1. Update relevant paths in transfer_learning.ipynb
  2. Run transfer_learning.ipynb

To load and use the pretrained model:

  • Run load_model.ipynb

To check inference rates of the model:

  • Run cal_inference.ipynb

For ways to improve model further please refer Improvement.md