Chessly (British Short Hairs) Classified as Persian Cat by ResNet-50

Chessly is one of my cat, she is a British shorthairs. However, she has been categorized as a Persian Cat by the ResNet-50 Image Classification Model.

Chessly has been classified as Persian by ResNet-50.

Introduction to ResNet-50

ResNet (Residual Network) was proposed by Microsoft Research in 2015. It solves the degradation problem in deep neural networks through residual structures.

ResNet-50 is one of the commonly used versions, with a total of 50 layers.

Core Idea

The residual block is computed as:

This shortcut connection helps avoid gradient vanishing and makes deeper networks easier to train.

Network Structure

The input image size is usually 224×224.

Main stages:

1. 7×7 convolution + max pooling
2. Conv2_x: 3 bottleneck blocks
3. Conv3_x: 4 bottleneck blocks
4. Conv4_x: 6 bottleneck blocks
5. Conv5_x: 3 bottleneck blocks
6. Global average pooling
7. Fully connected layer + Softmax classification

Key Features

• Easier to train with stable gradients
• Excellent performance on ImageNet (Top-5 error rate around 5%)
• About 25.5 million parameters
• Commonly used as backbone for detection and segmentation tasks

Applications

• Image classification tasks
• Feature extraction
• Transfer learning (medical imaging, satellite images, industrial inspection, etc.)

–EOF (The Ultimate Computing & Technology Blog) —