Stanford Car Classification

This is a project for the Stanford Car Classification Challenge on Kaggle. The goal is to classify cars into 196 classes.

💾 Folder data

The dataset contains 16,185 car images distributed over 196 classes/brands. There are 8,144 images for training and 8,041 images for testing in this dataset. Each class roughly has a 50-50 split in the training and validation set. The dataset is available here.

🚀 Folder models

This folder contains results observed from transfer-learning & fine-tuning the EfficientNet-b0 model on the Stanford Car dataset.

The models are trained using the following hyperparameters:

Classification Head	Optimizer	Learning Rate	Accuracy	Loss
Classification Head 1	Adam	lr=0.001	0.81	0.9
Classification Head 2	Adam	lr=0.001	0.81	0.8
Classification Head 3	Adam	lr=0.001	0.815	0.8
Classification Head 3	SGD	lr=0.001	0.82	0.6
Classification Head 3	Adam	StepLR (Decay every 1 epoch)	-	-
Classification Head 3	Adam	StepLR (Decay every 10 epoch)	-	-
Classification Head 3	Adam	ReduceLROnPlateau (factor=0.1)	-	-
Classification Head 3	SGD	StepLR (Decay every 10 epoch)	0.67	1.2
Classification Head 3	SGD	ReduceLROnPlateau (factor=0.1)	0.75	0.8
Classification Head 3	SGD	ReduceLROnPlateau (factor=0.5)	0.84	0.5

Classification Heads

The classification heads are the last layers of the model. The classification heads are trained on the top of the pre-trained model.

# Classficication Head 1
model.classifier[1] = nn.Linear(in_features=1280, out_features=num_classes).to(device)

# Classficication Head 2
model.classifier = torch.nn.Sequential(
  torch.nn.Linear(in_features=1280, out_features=640, bias=True),
  torch.nn.Dropout(p=0.2, inplace=True),
  torch.nn.Linear(in_features=640, out_features=320, bias=True),
  torch.nn.Dropout(p=0.2, inplace=True),
  torch.nn.Linear(in_features=320, out_features=num_classes, bias=True)).to(device)

# Classficication Head 3
model.classifier = nn.Sequential(
  nn.Linear(in_features=1280, out_features=640),
  nn.Linear(in_features=640, out_features=320),
  nn.Dropout(0.5),
  nn.Linear(in_features=320, out_features=num_classes)).to(device)

🔌 Folder src

Contains the source code for the project.

File	Description
dataset.py	Contains functions for loading and preprocess the data.
model.py	Contains the functions that define the model.
train.py	Contains the training and validation functions.
utils.py	Contains utility functions, such as plots and model saving.
main.py	Contains the main function that runs the training.

📝 File requirements.txt

Contains the list of dependencies for the project.

pip install -r requirements.txt