[Add] Initial Commit

Author: drawcodeboy

.gitignore

@@ -0,0 +1,16 @@
+# Virtual Environment
+.venv/
+
+# Cache
+**/__pycache__/**
+
+# VScode
+.vscode
+
+# Data
+data/*
+!data/temp.txt
+
+# Weights File
+saved/*
+!saved/temp.txt

README.md

@@ -0,0 +1,40 @@
+# PyTorch Project Template
+* 본 Repository는 PyTorch를 통해 프로젝트를 수행할 때 기반이 되는 코드베이스 템플릿입니다.
+
+# Installation
+```
+# Docker Container Setup
+docker pull ubuntu:22.04
+docker run -itd --gpus=all --shm-size=16G --name=<container_name> <image_name>
+
+# Ubuntu
+apt-get update
+apt-get install sudo
+sudo apt-get install python3
+python3 -m venv .venv
+source .venv/bin/activate
+```
+# Train & Test
+```
+# Train
+python train.py --config=convnet
+python train.py --config=convnet2
+
+# Test
+python test.py --config=convnet
+python test.py --config=convnet2
+```
+# Performances
+```
+# ConvNet (20 Epochs)
+Accuracy: 0.9540
+F1-Score(Macro): 0.9539
+Precision(Macro): 0.9572
+Recall(Macro): 0.9537
+
+# ConvNet2 (20 Epochs)
+Accuracy: 0.9700
+F1-Score(Macro): 0.9699
+Precision(Macro): 0.9701
+Recall(Macro): 0.9701
+```

configs/test.convnet.yaml

@@ -0,0 +1,22 @@
+expr: "Test ConvNet"
+
+device: cuda:0
+parallel: False
+
+data:
+  dataset: MNIST
+  root: data/
+  download: True
+  mode: test
+
+model:
+  name: ConvNet
+  in_channels: 1
+  layers: [8, 16, 32]
+  class_num: 10
+
+hyperparameters:
+  batch_size: 32
+
+save_dir: saved
+weights_file_name: convnet_mnist.epochs_100.pth

configs/test.convnet2.yaml

@@ -0,0 +1,22 @@
+expr: "Test ConvNet2"
+
+device: cuda:0
+parallel: False
+
+data:
+  dataset: MNIST
+  root: data/
+  download: True
+  mode: test
+
+model:
+  name: ConvNet2
+  in_channels: 1
+  layers: [8, 16, 32]
+  class_num: 10
+
+hyperparameters:
+  batch_size: 32
+
+save_dir: saved
+weights_file_name: convnet2_mnist.epochs_100.pth

configs/train.convnet.yaml

@@ -0,0 +1,26 @@
+expr: "Train ConvNet"
+
+device: cuda:0
+parallel: False
+
+data:
+  dataset: MNIST
+  root: data/
+  download: True
+  mode: train
+
+model:
+  name: ConvNet
+  in_channels: 1
+  layers: [8, 16, 32]
+  class_num: 10
+
+hyperparameters:
+  batch_size: 32
+  optim: AdamW
+  lr: 0.0001
+  loss_fn: cross-entropy
+  epochs: 100
+
+save_dir: saved
+save_name: convnet_mnist

configs/train.convnet2.yaml

@@ -0,0 +1,26 @@
+expr: "Train ConvNet2"
+
+device: cuda:0
+parallel: False
+
+data:
+  dataset: MNIST
+  root: data/
+  download: True
+  mode: train
+
+model:
+  name: ConvNet2
+  in_channels: 1
+  layers: [8, 16, 32]
+  class_num: 10
+
+hyperparameters:
+  batch_size: 32
+  optim: AdamW
+  lr: 0.0001
+  loss_fn: cross-entropy
+  epochs: 100
+
+save_dir: saved
+save_name: convnet2_mnist

data/temp.txt

@@ -0,0 +1 @@
+should include this directory

datasets/__init__.py

@@ -0,0 +1,10 @@
+from .mnist_dataset import MNIST_Dataset
+
+def load_dataset(**cfg):
+    if cfg['dataset'] == 'MNIST':
+        return MNIST_Dataset(root=cfg['root'],
+                             download=cfg['download'],
+                             mode=cfg['mode'])
+    
+    else:
+        raise Exception(f"Dataset: {cfg['dataset']} is not supported.")

datasets/mnist_dataset.py

@@ -0,0 +1,40 @@
+import torch
+from torch.utils.data import dataset
+from torchvision.datasets import MNIST
+import numpy as np
+
+class MNIST_Dataset():
+    def __init__(self,
+                 root="data/", # make MNIST directory below data/
+                 download=True, # download if there is no data, else pass
+                 mode='train'): 
+        
+        if mode not in ['train', 'test']:
+            raise AssertionError(f"Mode-related Error: [{mode}] is not supported.")
+        
+        self.data = MNIST(root=root,
+                    download=download)
+        
+        self.data=list(self.data)
+        
+        train_size = 5000
+        test_size = 500
+        
+        if mode == 'train':
+            self.data = self.data[:train_size]
+        elif mode == 'test':
+            self.data = self.data[train_size:train_size+test_size]
+        
+    def __len__(self):
+        return len(self.data)
+    
+    def __getitem__(self, idx):
+        image, label = self.data[idx]
+        
+        image = np.array(image.getdata()).reshape(28, 28).astype(np.float32)
+        image /= 255.
+        image = torch.tensor(image, dtype=torch.float32).unsqueeze(0)
+        
+        label = torch.tensor(label, dtype=torch.int64)
+        
+        return image, label

models/ConvNet/block.py

@@ -0,0 +1,28 @@
+from torch import nn
+
+class Block(nn.Module):
+    def __init__(self,
+                 in_dim:int = 16,
+                 out_dim:int = 32):
+        
+        super().__init__()
+        
+        # Convolution layer
+        self.conv1 = nn.Conv2d(in_dim, out_dim, kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2d(out_dim, out_dim, kernel_size=3, stride=1, padding=1)
+        
+        # Normalization layer
+        self.bn1 = nn.BatchNorm2d(out_dim)
+        self.bn2 = nn.BatchNorm2d(out_dim)
+        
+        # Activation funciton
+        self.relu = nn.ReLU()
+        
+    def forward(self, x):
+        x = self.bn1(self.conv1(x))
+        x = self.relu(x)
+        
+        x = self.bn2(self.conv2(x))
+        x = self.relu(x)
+        
+        return x