calority_scratch_model.py

import json
from pathlib import Path

import torch
from PIL import Image
from torch import nn


IMAGE_SIZE = 224
MEAN = torch.tensor([0.485, 0.456, 0.406]).view(3, 1, 1)
STD = torch.tensor([0.229, 0.224, 0.225]).view(3, 1, 1)


class CalorityFoodCNN(nn.Module):
    def __init__(self, num_labels: int):
        super().__init__()
        self.features = nn.Sequential(
            self._block(3, 32),
            self._block(32, 64),
            self._block(64, 128),
            self._block(128, 256),
            self._block(256, 384),
        )
        self.pool = nn.AdaptiveAvgPool2d((1, 1))
        self.classifier = nn.Sequential(
            nn.Flatten(),
            nn.Dropout(0.35),
            nn.Linear(384, 256),
            nn.ReLU(inplace=True),
            nn.Dropout(0.2),
            nn.Linear(256, num_labels),
        )

    @staticmethod
    def _block(in_channels: int, out_channels: int) -> nn.Sequential:
        return nn.Sequential(
            nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(2),
        )

    def forward(self, pixel_values: torch.Tensor) -> torch.Tensor:
        x = self.features(pixel_values)
        x = self.pool(x)
        return self.classifier(x)


def image_to_tensor(image: Image.Image, image_size: int = IMAGE_SIZE) -> torch.Tensor:
    resized = image.convert("RGB").resize((image_size, image_size), Image.Resampling.BILINEAR)
    raw = torch.ByteTensor(torch.ByteStorage.from_buffer(resized.tobytes()))
    tensor = raw.view(image_size, image_size, 3).permute(2, 0, 1).float() / 255.0
    return (tensor - MEAN) / STD


def save_checkpoint(model: nn.Module, labels: list[str], output_dir: str | Path) -> None:
    output_path = Path(output_dir)
    output_path.mkdir(parents=True, exist_ok=True)
    torch.save(model.state_dict(), output_path / "model.pt")
    (output_path / "labels.json").write_text(json.dumps(labels, indent=2), encoding="utf-8")
    (output_path / "config.json").write_text(
        json.dumps({"architecture": "CalorityFoodCNN", "image_size": IMAGE_SIZE}, indent=2),
        encoding="utf-8",
    )


def load_checkpoint(model_dir: str | Path, device: str | torch.device = "cpu") -> tuple[CalorityFoodCNN, list[str]]:
    model_path = Path(model_dir)
    labels = json.loads((model_path / "labels.json").read_text(encoding="utf-8"))
    model = CalorityFoodCNN(num_labels=len(labels))
    state = torch.load(model_path / "model.pt", map_location=device)
    model.load_state_dict(state)
    model.to(device)
    model.eval()
    return model, labels