TF Lite
sgtsaughter's picture
Update README.md
2a2d8b4 verified
|
raw
history blame
684 Bytes
metadata
license: apache-2.0
datasets:
  - sgtsaughter/pokemon-classification-images-151

The model was trained from this data set sgtsaughter/pokemon-classification-images-151

I trained this model using google colab pro account using T4 GPU and High-RAM turned on (Go to Runtime -> Change runtime type in google colab to update this). The training took about 30 to 40 minutes.
It uses MobileNetV2 as the base model

The .ipynb file assumes you're connecting your google drive, and in that drive had a folder called Dataset with a PokemonData.zip file within it.

The model uses early stopping to prevent overfitting.

It then goes through another round of fine-tuning afterwards.