CIFAR-10 Image Classifier
Test my custom-trained Convolutional Neural Network directly in your browser. Upload any photo, drag & drop, paste from clipboard, or click a sample below to see real-time inference with zero server lag.
Input Image
Upload a photo or choose an authentic CIFAR-10 test sample.
Drop an image here, or click to browse
Supports JPG, PNG, WebP • Or press Ctrl+V to paste
Click a sample image:
CIFAR-10 Test Set
Original Image
128 × 128 px
What the CNN Sees (32×32)
32 × 32 px (Normalized RGB)
Predictions
Softmax probability distribution across all 10 CIFAR-10 classes.
✈️
Airplane
Inference: 12 ms94.8%
About This CNN Research Project
This Convolutional Neural Network was designed, tuned, and trained from scratch in Python with TensorFlow/Keras on 60,000 32x32 labeled images from the CIFAR-10 dataset (50,000 training, 10,000 testing). It was exported directly to ONNX format to execute natively inside your browser using WebAssembly and WebGL.
- Architecture: 3 Convolutional Blocks (Conv2D 32 → 64 → 128 filters with 3x3 kernels), each paired with Batch Normalization, ReLU activations, and 2x2 Max Pooling.
- Classifier Head: Flatten → Dropout (0.5) → Dense (64 units) → Dropout (0.5) → Dense (10 units, Softmax). Totaling 225,930 parameters (~880 KB).
- Performance Tradeoffs: CIFAR-10 images are only 32x32 pixels. When testing with your own photos, images with centered subjects and simple backgrounds achieve the most accurate classifications.