Computer Vision
Predict the Product Category of the Image
Analysis Results
Analyzed Image
Predictions
Computer Vision System
A sophisticated computer vision system leveraging deep learning and transfer learning techniques to perform accurate product image classification. The system utilizes MobileNetV2 architecture with custom modifications for optimal performance on e-commerce product images.
Model Architecture
- Base Model: MobileNetV2 (pre-trained on ImageNet, frozen layers)
-
Custom Layers:
- • Flatten
- • Dense (512) + ReLU + BatchNorm + Dropout(0.5)
- • Dense (256) + ReLU + BatchNorm + Dropout(0.5)
- • Dense (10) + Softmax
- Input Shape: 224x224x3 (RGB)
Training Configuration
Batch Size: 32
Epochs: 20
Early Stopping (patience=5)
LR Reduction (factor=0.1)
Loss: Categorical Crossentropy
Optimizer: Adam (lr=1e-4)
Data Pipeline
Training Augmentation
- Rescale: 1/255
- Horizontal Flip
- Zoom Range: 20%
- Shear Range: 20%
- Batch Size: 32
- Shuffle: Enabled
Validation Pipeline
- Rescale: 1/255
- No Augmentation
- Batch Size: 32
- Shuffle: Disabled
Available Optimizers
SGD
lr=1e-3, momentum=0.9, nesterov=True
RMSprop
lr=1e-4
Adam
lr=1e-4 (Selected)
AdamW
lr=1e-4, weight_decay=1e-5
Adagrad
lr=1e-4
Adadelta
lr=1e-4
Adamax
lr=1e-4
Nadam
lr=1e-4
Performance Metrics
81%
Overall Accuracy
0.80
Weighted F1-Score
Category Performance
Top Categories (F1):
- • Clothing/Accessories: 0.93
- • Baby Products: 0.90
- • Grocery: 0.89
Precision Range:
- • Highest: 1.00 (Medicine)
- • Average: 0.83
- • Lowest: 0.67 (Home/Kitchen)
Detailed Metrics
| Category | Precision | Recall | F1-Score |
|---|---|---|---|
| Baby Products | 0.91 | 0.89 | 0.90 |
| Beauty & Health | 0.72 | 0.66 | 0.69 |
| Clothing & Accessories | 0.96 | 0.89 | 0.93 |
| Electronics | 0.74 | 0.87 | 0.80 |
| Grocery | 0.85 | 0.94 | 0.89 |
| Hobby & Arts | 0.81 | 0.71 | 0.76 |
| Home & Kitchen | 0.67 | 0.61 | 0.64 |
| Medicine & Supplements | 1.00 | 0.20 | 0.33 |
| Pet Supplies | 0.87 | 0.72 | 0.79 |
| Sports & Outdoor | 0.73 | 0.78 | 0.75 |
Technologies Used
NumPy
Numerical Computing
OpenCV
Image Processing
scikit-learn
ML Utilities
MobileNetV2
Base Architecture
PIL
Image Handling
ImageNet
Pre-training