Does Mimi Need A Haircut?

Deep Learning Computer Vision Transfer Learning Multi-task Learning Streamlit

Architecture

Does Mimi Need A Haircut? pipeline diagram

An end-to-end computer vision pipeline that predicts coat length and grooming needs from photographs of Mimi, a dog. The project covers the full ML lifecycle: image ingestion, EXIF-based date structuring, manual CLI labelling, data augmentation, multi-task training, and a live Streamlit deployment.

A fine-tuned MobileNetV2 backbone drives four simultaneous output heads: hair length class (S/M/L), coat length in millimetres, coat texture, and camera view angle. Splits are date-based to prevent temporal leakage across Mimi's grooming history.

The Streamlit inference app augments predictions with live weather context from Open-Meteo to give grooming advice relevant to current conditions.

Training results (71 labelled images, 12-image test set): hair length accuracy 41.7% · coat length MAE 2.74 mm. Dataset is early-stage; performance is expected to improve substantially with a larger labelled set.

Built with

Python TensorFlow Keras MobileNetV2 Streamlit Pillow Pandas

Multi-output transfer learning on a custom-labelled 71-image dataset; 4 simultaneous heads (classification ×3, regression ×1); date-based train/val/test splits to prevent temporal leakage.

← UK Railway Analytics Next: MoE Emotion Predictor →