Visual Intelligence & Deep Learning Oct 11, 2025 Published project
Animal Faces CNN Classification

Image classification workflow with balanced visual evaluation

This project trains a CNN classifier for animal-face recognition across Cat, Dog, and Wild classes. It focuses on image preprocessing, augmentation, callback-based training, and class-level performance review.

View source code Back to Projects
PythonTensorFlowKerasCNNData AugmentationCallbacks
Share project X LinkedIn Email

Challenge

  • Visual classes can overlap because animal faces have high intra-class variation.
  • Training needs augmentation and callbacks to reduce overfitting and stabilize learning.
  • Evaluation should inspect per-class behavior, not only one headline accuracy.

System architecture

Image datasetCat, Dog, Wild
CNN feature learningConvolutional blocks
Training controlCallbacks and checkpoints
EvaluationConfusion matrix and samples

Data and inputs

  • Animal Faces dataset with Cat, Dog, and Wild categories.
  • 128×128 RGB inputs with pixel values normalized to [0, 1].
  • 11,706 training images and 2,924 test/validation images.

Technical approach

  • Use stacked convolutional blocks with max pooling, batch normalization, and dropout.
  • Train with Adam, early stopping, learning-rate reduction, and checkpointing.
  • Review training curves, confusion matrix, and prediction examples.

Evaluation and results

Key indicators

11,706 training images

Key indicators

2,924 test/validation images

Key indicators

Overall accuracy 0.9593

  • The CNN achieved 0.9593 overall accuracy across the three classes.
  • Cat classification was strongest; Wild recall was lower due to greater visual diversity.
  • Training and validation curves showed stable convergence without severe overfitting.

Implementation and code

Implementation focus

The implementation connects data preparation, modeling, evaluation, and interpretation in a structured workflow that makes the technical decisions clear.

Source code

The code is available for exploring the implementation details and extending the experiment when needed.

Open source code

Scope and responsible use

The project is a focused modeling and evaluation study. Broader use should be supported by validation on additional data, robustness checks, monitoring, and domain-specific evaluation.

Future development

  • Add stronger augmentation policies and transfer-learning baselines.
  • Inspect failure cases by species and image quality.
  • Package the classifier into a small demo interface.

Technical contribution

The project demonstrates end-to-end visual classification: preprocessing, training control, class-level metrics, and visual error inspection.