Image Classification

import torch
from torchvision import models, transforms
from PIL import Image

# ── Load ResNet-50 đã pretrain trên ImageNet ────────────────
model = models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V2)
model.eval()

# ── Tiền xử lý: chuẩn theo ImageNet stats ──────────────────
transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225],
    ),
])

img = Image.open("xe_may.jpg").convert("RGB")
x = transform(img).unsqueeze(0)          # (1, 3, 224, 224)

with torch.no_grad():
    logits = model(x)                    # (1, 1000)
    probs  = torch.softmax(logits, dim=1)
    top5   = torch.topk(probs, 5)

# ── In ra Top-5 cùng xác suất ──────────────────────────────
classes = models.ResNet50_Weights.IMAGENET1K_V2.meta["categories"]
for i in range(5):
    idx = top5.indices[0][i].item()
    p   = top5.values[0][i].item()
    print(f"{i+1}. {classes[idx]:30s}  {p*100:5.2f}%")

# Output ví dụ:
# 1. moped                          82.41%
# 2. mountain_bike                   7.12%
# 3. motor_scooter                   4.88%
# 4. unicycle                        2.05%
# 5. tricycle                        1.33%

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torchvision import models, datasets, transforms

# ── 1) Dataset: ảnh sản phẩm Shopee VN, 10 lớp ─────────────
train_tf = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.ColorJitter(0.3, 0.3, 0.3),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])
train_ds = datasets.ImageFolder("data/shopee_vn/train", transform=train_tf)
train_ld = DataLoader(train_ds, batch_size=64, shuffle=True, num_workers=4)

# ── 2) Load pretrained, thay lớp cuối ──────────────────────
model = models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V2)
num_classes = 10
model.fc = nn.Linear(model.fc.in_features, num_classes)
model = model.cuda()

# ── 3) Fine-tune toàn bộ với LR nhỏ cho backbone ──────────
optimizer = torch.optim.AdamW([
    {"params": [p for n, p in model.named_parameters() if "fc" not in n], "lr": 1e-4},
    {"params": model.fc.parameters(), "lr": 1e-3},   # head LR lớn hơn
], weight_decay=1e-4)

criterion = nn.CrossEntropyLoss(label_smoothing=0.1)
scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=20)

# ── 4) Vòng train ngắn gọn ────────────────────────────────
for epoch in range(20):
    model.train()
    for x, y in train_ld:
        x, y = x.cuda(), y.cuda()
        optimizer.zero_grad()
        loss = criterion(model(x), y)
        loss.backward()
        optimizer.step()
    scheduler.step()
    print(f"epoch {epoch+1:02d}  loss={loss.item():.4f}")