dlgenai-nppe

Age and Gender Prediction Model

This is a model created for the "Sep 25 DLGenAI NPPE 1" Kaggle competition.

🚀 Project Goal

The goal is to predict a person's Age (regression) and Gender (binary classification) from a facial image.

🧠 Model Details

This repository contains a finetuned ResNet50 model.

• Backbone: ResNet50 (pretrained on ImageNet)
• Finetuning: The last two blocks (layer3 and layer4) were unfrozen and trained with a low learning rate.
• Heads: Two separate classification/regression heads were added for the age and gender tasks.
• Loss: A weighted loss was used to balance the two tasks: total_loss = loss_age + (20 * loss_gender)

🏆 Performance

• Score V1 (ScratchCNN): 0.45
• Score V2 (ResNet50 Frozen): 0.71
• Score V3 (ResNet50 Finetuned): TBD (Best val_loss: 11.038)

💻 Usage

This model is a pytorch_lightning checkpoint. It can be loaded as follows:

from huggingface_hub import hf_hub_download
import pytorch_lightning as pl

# Make sure the ResNet50FinetuneModel class is defined
model_path = hf_hub_download(repo_id="MayKeeGarmi/dlgenai-nppe", filename="model.ckpt")
model = ResNet50FinetuneModel.load_from_checkpoint(model_path)