Initially could just be an embedding layer or two added as input to the current modeling.models.EfficientNet code, passing the embedding up and concatenating with the EfficientNet output to form the final representation before going into the final Dense layer. Could eventually be something more advanced.
Initially could just be an embedding layer or two added as input to the current modeling.models.EfficientNet code, passing the embedding up and concatenating with the EfficientNet output to form the final representation before going into the final Dense layer. Could eventually be something more advanced.