howest-hf-ai: cat, dog or panda

Small CNN (5 conv blocks + dense layer, about 4.6M parameters) that classifies a 64x64 RGB image as cat, dog or panda.

• Trained in an Azure ML pipeline (CLI, YAML) for 5 epochs, SGD, batch size 32, data from the Howest course (1000 images per class, 80/20 split).
• Test accuracy: 0.69 (cats recall 0.41, dogs 0.77, pandas 0.90). It is trained for only a few epochs on a small dataset, so it is not robust.
• Label order: cat, dog, panda (index 0, 1, 2). Input: uint8 image resized to 64x64, no extra scaling.
• Files are the unpacked Keras format (config.json, metadata.json, model.weights.h5).

import os
os.environ["KERAS_BACKEND"] = "tensorflow"
import keras
model = keras.saving.load_model("hf://MoorsArthur/howest-hf-ai")

Made for the Howest MLOps course, assignment 1.3.