Introduction#
这次的 HW 是图像分类任务，对食物数据集 food11 进行分类，具体的任务要求如下：
需要达到的目标和 Hints 分别为：
Simple#
只需要 run 它的原始代码即可，结果如下：
Medium#
进行 Training Augmentation 并将训练时间延长（用更大的 n_epoch）。
Training Augmentation 具体操作如下：
train_tfm = transforms.Compose([
# Resize the image into a fixed shape (height = width = 128)
transforms.Resize((128, 128)),
# You may add some transforms here.
transforms.RandomChoice([
transforms.RandomRotation((-30,30)),
transforms.RandomHorizontalFlip(p=0.5),
transforms.RandomVerticalFlip(p=0.5),
transforms.ColorJitter(brightness=(0.5,1.5), contrast=(0.5, 1.5), saturation=(0.5,1.5), hue=(-0.25, 0.25)),
transforms.RandomInvert(p=0.5),
transforms.RandomAffine(degrees=(-30,30), translate=(0.1, 0.1), scale=(0.8, 1.2), shear=(-30, 30)),
transforms.Grayscale(num_output_channels=3),
]),
# ToTensor() should be the last one of the transforms.
transforms.ToTensor(),
])
具体介绍如下：
- RandomRotation ((-30,30)) 随机旋转图像。(-30, 30)：旋转角度范围（-30 度到 + 30 度之间随机选择）。
- RandomHorizontalFlip (p=0.5)，以 50% 概率水平翻转图像。p=0.5：执行概率（0.5 表示 50%）。
- RandomVerticalFlip (p=0.5)，以 50% 概率垂直翻转图像。p 含义同上。
- ColorJitter (brightness=(0.5,1.5), contrast=(0.5, 1.5), saturation=(0.5,1.5), hue=(-0.25, 0.25))，随机调整颜色属性（亮度、对比度、饱和度、色调）。brightness=(0.5, 1.5)：亮度缩放范围（0.5 倍～1.5 倍）。contrast=(0.5, 1.5)：对比度调整范围（0.5 倍～1.5 倍）。saturation=(0.5, 1.5)：饱和度调整范围（0.5 倍～1.5 倍）。hue=(-0.25, 0.25)：色调偏移范围（-0.25~+0.25，对应色相环的 - 90 度～+90 度）。
- RandomInvert (p=0.5)，以 50% 概率反色（颜色取反，如黑变白、红变青）。p 含义同上。
- RandomAffine (degrees=(-30,30), translate=(0.1, 0.1), scale=(0.8, 1.2), shear=(-30, 30))，随机仿射变换（旋转、平移、缩放、剪切）。degrees=(-30, 30)：旋转角度范围。translate=(0.1, 0.1)：水平和垂直方向最大平移比例（10% 图像尺寸）。scale=(0.8, 1.2)：缩放范围（0.8 倍～1.2 倍）。shear=(-30, 30)：剪切角度范围（-30 度～+30 度）。剪切（Shear）是一种线性几何变换，通过倾斜图像的一部分来模拟 “倾斜变形” 效果。
- Grayscale (num_output_channels=3)，将图像转为灰度图，但保留 3 通道（RGB 格式，每通道值相同）。num_output_channels=3：输出通道数（3 表示生成 3 通道灰度图，兼容模型输入）。将彩色图像（RGB）转换为灰度图，本质是通过加权平均合并三个通道的亮度信息，生成单通道图像。
Training Time 延长到 90 个 epoch，结果如下：
需要注意的是，查看 gpu 利用率发现 gpu 利用率很低，问题应该是出在每次读取数据时候的 tranform 增强还有磁盘 IO 的时间太长了，导致训练时间很慢，需要调一下 DataLoader 的并发。
Train 此处使用了 12 个并发（取决于图像增强的复杂度），取 persistent_workers=True，避免重复创建 / 销毁进程，减少开销。Test 使用了 8 个并发。改善后训练一个 epoch 的时间大大缩短，此处使用的是本地电脑跑的，设备为 RTX4070laptop。
同时还将 batchsize 变成了 128。
Strong#
首先先将模型结构进行了修改，学习了一下 resnet18 和 resnet34，自己在 resnet34 的基础上小改了一下：
class BasicBlock(nn.Module):
def __init__(self, in_channels, out_channels, stride=1):
super().__init__()
self.conv1 = nn.Conv2d(in_channels, out_channels, 3, stride, 1, bias=False)
self.bn1 = nn.BatchNorm2d(out_channels)
self.conv2 = nn.Conv2d(out_channels, out_channels, 3, 1, 1, bias=False)
self.bn2 = nn.BatchNorm2d(out_channels)
self.relu = nn.ReLU(inplace=True)
self.downsample = None
if stride != 1 or in_channels != out_channels:
self.downsample = nn.Sequential(
nn.Conv2d(in_channels, out_channels, 1, stride, 0),
nn.BatchNorm2d(out_channels)
)
def forward(self, x):
identity = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
if self.downsample is not None:
identity = self.downsample(identity)
out += identity
out = self.relu(out)
return out
class Classifier(nn.Module):
def __init__(self):
super(Classifier, self).__init__()
# torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding)
# torch.nn.MaxPool2d(kernel_size, stride, padding)
# input 維度 [3, 128, 128]
# 初始卷积层
self.conv1 = nn.Conv2d(3, 64, kernel_size=5, stride=2, padding=2, bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.block1 = BasicBlock(64, 64, 1) # [64, 64, 64]
self.block2 = BasicBlock(64, 64) # [64, 64, 64]
self.block3 = BasicBlock(64, 64) # [64, 64, 64]
self.block4 = BasicBlock(64, 128, 2) # [128, 32, 32]
self.block5 = BasicBlock(128, 128) # [128, 32, 32]
self.block6 = BasicBlock(128, 128) # [128, 32, 32]
self.block7 = BasicBlock(128, 128) # [128, 32, 32]
self.block8 = BasicBlock(128, 256, 2) # [256, 16, 16]
self.block9 = BasicBlock(256, 256) # [256, 16, 16]
self.block10 = BasicBlock(256, 256) # [256, 16, 16]
self.block11 = BasicBlock(256, 256) # [256, 16, 16]
self.block12 = BasicBlock(256, 256) # [256, 16, 16]
self.block13 = BasicBlock(256, 256) # [256, 16, 16]
self.block14 = BasicBlock(256, 512, 2) # [512, 8, 8]
self.block15 = BasicBlock(512, 512) # [512, 8, 8]
self.block16 = BasicBlock(512, 512) # [512, 8, 8]
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512, 11)
def forward(self, x):
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.block1(out)
out = self.block2(out)
out = self.block3(out)
out = self.block4(out)
out = self.block5(out)
out = self.block6(out)
out = self.block7(out)
out = self.block8(out)
out = self.block9(out)
out = self.block10(out)
out = self.block11(out)
out = self.block12(out)
out = self.block13(out)
out = self.block14(out)
out = self.block15(out)
out = self.block16(out)
out = self.avgpool(out)
out = out.view(out.size()[0], -1)
return self.fc(out)
然后进行了 Cross-Validation 和 Ensemble：
Cross-Validation 采用了五折交叉验证，将原来的训练集和验证集合并后再进行的五折交叉验证：
# "cuda" only when GPUs are available.
device = "cuda" if torch.cuda.is_available() else "cpu"
# The number of training epochs and patience.
n_epochs = 200
patience = 50 # If no improvement in 'patience' epochs, early stop
import numpy as np
from sklearn.model_selection import KFold
from torch.utils.tensorboard import SummaryWriter
import datetime
# 初始化5折交叉验证
n_folds = 5
kf = KFold(n_splits=n_folds, shuffle=True, random_state=42)
# 加载完整训练集（用于交叉验证）
train_set = FoodDataset(os.path.join(_dataset_dir, "training"), tfm=train_tfm)
valid_set = FoodDataset(os.path.join(_dataset_dir, "validation"), tfm=train_tfm)
# 合并数据集
combined_files = train_set.files + valid_set.files
full_dataset = FoodDataset(path="", tfm=train_tfm, files=combined_files)
oof_preds = np.zeros(len(full_dataset)) # 存储OOF预测结果
oof_labels = np.zeros(len(full_dataset)) # 存储真实标签
# 存储所有基模型（用于后续集成）
base_models = []
timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
log_dir = f"runs/food_classification_{timestamp}"
writer = SummaryWriter()
for fold, (train_idx, val_idx) in enumerate(kf.split(train_set)):
print(f"\n====== Fold {fold+1}/{n_folds} ======")
# 划分训练集和验证子集
train_subset = Subset(train_set, train_idx)
val_subset = Subset(train_set, val_idx)
# DataLoader
train_loader = DataLoader(
train_subset,
batch_size=batch_size,
shuffle=True,
num_workers=12,
pin_memory=True,
persistent_workers=True
)
val_loader = DataLoader(
val_subset,
batch_size=batch_size,
shuffle=False,
num_workers=8,
pin_memory=True,
persistent_workers=True
)
# 每折独立初始化模型和优化器
model = Classifier().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.0003, weight_decay=1e-5)
criterion = nn.CrossEntropyLoss()
# 早停相关变量（每折独立）
fold_best_acc = 0
stale = 0
# 训练循环（保持原有逻辑）
for epoch in range(n_epochs):
# ---------- Training ----------
model.train()
train_loss, train_accs = [], []
for batch in tqdm(train_loader, desc=f"Epoch {epoch+1}"):
imgs, labels = batch
imgs, labels = imgs.to(device), labels.to(device)
logits = model(imgs)
loss = criterion(logits, labels)
optimizer.zero_grad()
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10)
optimizer.step()
acc = (logits.argmax(dim=-1) == labels).float().mean()
train_loss.append(loss.item())
train_accs.append(acc.item())
# 打印训练信息
avg_loss = np.mean(train_loss)
avg_acc = np.mean(train_accs)
# 写入TensorBoard
writer.add_scalar(f'Fold_{fold}/Train/Loss', avg_loss, epoch)
writer.add_scalar(f'Fold_{fold}/Train/Accuracy', avg_acc, epoch)
print(f"[ Train | {epoch+1:03d}/{n_epochs:03d} ] loss = {avg_loss:.5f}, acc = {avg_acc:.5f}")
# ---------- Validation ----------
model.eval()
val_loss, val_accs, val_preds = [], [], []
val_labels = [] # 累积所有验证批次的标签
for batch in tqdm(val_loader, desc="Validating"):
imgs, labels = batch
imgs = imgs.to(device)
labels_np = labels.numpy()
val_labels.extend(labels_np) # 累积标签
with torch.no_grad():
logits = model(imgs)
preds = logits.argmax(dim=-1).cpu().numpy()
loss = criterion(logits, labels.to(device))
val_loss.append(loss.item())
val_accs.append((preds == labels_np).mean())
val_preds.extend(preds)
# 记录OOF预测和标签
oof_preds[val_idx] = np.array(val_preds)
oof_labels[val_idx] = np.array(val_labels)
# 打印验证信息
avg_val_loss = np.mean(val_loss)
avg_val_acc = np.mean(val_accs)
# 写入TensorBoard
writer.add_scalar(f'Fold_{fold}/Val/Loss', avg_val_loss, epoch)
writer.add_scalar(f'Fold_{fold}/Val/Accuracy', avg_val_acc, epoch)
print(f"[ Valid | {epoch+1:03d}/{n_epochs:03d} ] loss = {avg_val_loss:.5f}, acc = {avg_val_acc:.5f}")
# 早停逻辑（每折独立）
if avg_val_acc > fold_best_acc:
print(f"Fold {fold} best model at epoch {epoch}")
torch.save(model.state_dict(), f"fold{fold}_best.ckpt")
fold_best_acc = avg_val_acc
stale = 0
else:
stale += 1
if stale > patience:
print(f"Early stopping at epoch {epoch}")
break
# 保存当前折的模型
base_models.append(model)
# 关闭TensorBoard的writer
writer.close()
# ---------- 后处理 ----------
# 计算OOF准确率
oof_acc = (oof_preds == oof_labels).mean()
print(f"\n[OOF Accuracy] {oof_acc:.4f}")
保存五个基模型后在 test 部分使用 ensemble：
# 集成预测（软投票法）
all_preds = []
for model in base_models:
model.eval()
fold_preds = []
for data, _ in test_loader: # 保持和原test_loader一致
with torch.no_grad():
logits = model(data.to(device))
# 保存每个模型的原始logits（概率），而不是直接argmax
fold_preds.append(logits.cpu().numpy())
# 合并当前模型的所有batch预测结果
fold_preds = np.concatenate(fold_preds, axis=0)
all_preds.append(fold_preds)
# 软投票：平均所有模型的logits后取argmax
all_preds = np.stack(all_preds) # shape: (n_models, n_samples, n_classes)
prediction = all_preds.mean(axis=0).argmax(axis=1) # shape: (n_samples,)
最后结果如下：
已经非常接近 boss 了
由于时间问题就没有做 boss，以后有时间再回来补吧。
两个 report_problem 分别是数据增强和设计残差网络，在 medium 和 strong 完成的过程里有，就没有额外再说了。