JunrQ / celebA

Project in Advanced Computer Vision course in NTU

AI6126 Project 1: CelebA Facial Attribute Recognition Challenge

Required packages

• pytorch-lighting
• pytorch
• pandas = 1.1.1 required by dataset.CelebA
• requests = 2.24.0
• tqdm = 4.49.0

File descriptions

• models.py : Define the model architeture and training, evaluation operation.
• trainer.py : Define trainer.
• nets.py : Some backbones.
• run.py : Script used for running the training and evaluation.
• utils.py : Define some functions.

Run

In order to train or test a model, using the following script

# Specify the configuration used
config=v1.0.1.config

# Specify gpus
gpus=0

# Run
python -u run.py \
  --config config/${config} \
  --path log/${config} \
  --gpus ${gpus} \
  > log/${config}.log 2>&1

Experiments Results

version	Test Accuracy
v0.1.config	0.9058
v0.6.config	0.9064
v0.7.config	0.9100
v0.7.1.config	0.9115
v0.7.2.config	0.9172
v0.7.3.config	0.9173
v1.0.0.config	0.9043