Build a Shiba‑Inu vs Akita Detector with Keras on Alibaba Cloud Serverless AI

As a fan of Shiba Inu videos, the author often confuses Shiba Inu with Akita and discovered an Alibaba Cloud developer experiment that builds a TensorFlow serverless AI inference platform for cat‑dog classification.

By swapping the original model with one trained on Shiba Inu and Akita images, a "Shiba‑Inu vs Akita Detector" can be created.

Keras, a high‑level Python neural‑network API, is used with TensorFlow as the backend to construct a convolutional neural network (CNN) suitable for image‑recognition tasks.

First, 500 Shiba Inu and 500 Akita pictures are collected and labeled. Then a three‑layer CNN is built:

Convolution layer : applies filters to extract basic visual features.

Pooling layer : reduces dimensionality while preserving key features.

Fully‑connected layer : performs classification to produce the final prediction.

The model implementation follows code from a Kaggle notebook (https://www.kaggle.com/uysimty/keras-cnn-dog-or-cat-classification). After training with an 80/20 split for training and validation, the model is saved.

Deployment steps:

Clone the project locally.

Replace the models folder with the trained Shiba‑Inu/Akita model.

Modify index.html and predict.py to change references from cats/dogs to Shiba‑Inu/Akita.

The service is deployed on Alibaba Cloud Function Compute, a serverless environment where code and dependencies are packaged and deployed with a single command.

Alibaba Cloud provides the Funcraft tool, similar to Docker, Kubernetes, or Terraform, allowing users to describe all resources (functions, API gateways, log services, etc.) in a template file and deploy them via CLI.

After deployment, the detector is tested with sample Shiba Inu and Akita images; accuracy is acceptable but not perfect, suggesting room for improvement.

Beyond this experiment, Alibaba Cloud’s developer platform offers many other scenarios (e.g., cloud databases, ECS) for beginners to practice and obtain free cloud resources.