You're seeing a computer vision test field, under the presented architecture, you can easily try some architecture such as AlexNet, VGGx16, VGGx19 and ResNetx18
python -m venv env
env\Scripts\activate
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install numpy matplotlib python-dotenv plyer
python -m venv env
source env/bin/activate
pip install torch torchvision torchaudio
pip install numpy matplotlib python-dotenv plyer
.env file is required to setting the environment of this project, along with the virtual environment this file sets each section of the project. Here is a example
# Architecture
NET_ARCH=AlexNet
USE_CUDA=1
# Using model
MODELS_PATH="models"
USE_MODEL=1
# Dataset
DATASET="CIFAR10"
DATA_PATH="./data"
BATCH_SIZE=8
# Image management
IMG_SIZE=224
IMG_START_INDEX=0
# Training
ITERATIONS=1
LEARNING_RATE=0.01
MOMENTUM_VALUE=0.8
CATCH_INTERVAL=5
# Loss
LOST_CRITERIA="CrossEntropyLoss"
# Management
RESULTS_PATH="results"
LOG_PATH="log"
AUTOCLEAR=0
Just write any of the following on the NET_ARCH env var
- AlexNet
- VGG16
- VGG19
- ResNet
USE_CUDA=1 means that the host can and will use CUDA by default it uses the processor
Data sets can be defined inside the .env file in the
Keyword | Size | Dataset |
---|---|---|
"CelebA" | 200K | CelebA |
"CIFAR10" | 60K | CIFAR |
"CIFAR100" | 60K | CIFAR |
Note: The difference between CIFAR10 and CIFAR100 is the amount of classes, CIFAR10 contains 10 while CIFAR100 contains 100 see "The CIFAR-100 dataset" specifications
LOST_CRITERIA
means the lost function, available options are "BCELoss" and "CrossEntropyLoss"