embeddings.py

"""
This module provides the embedding store and the methods to add embeddings to
it from images.
"""

import operator
import pathlib
from typing import Any, Dict, Iterable, List, Optional, Tuple

import h5py
import numpy as np
import torch
from efficientnet_pytorch import EfficientNet
from PIL import Image

import settings


class EmbeddingStore:
    """A class to store logo data and embeddings.
    
    Embeddings are stored locally on disk in an HDF5 file.

    :param hdf5_path: Path of the HDF5 file where the logo embeddings are
    stored. If the file does not exist, the file will be created the first
    time `save_embeddings` is called
    """

    def __init__(self, hdf5_path: pathlib.Path):
        self.hdf5_path = hdf5_path
        self.logo_id_to_idx: Dict[int, int] = self.load_logo_id_to_index(hdf5_path)
        self.offset = (
            max(self.logo_id_to_idx.values()) + 1 if self.logo_id_to_idx else 0
        )

    def __len__(self):
        return len(self.logo_id_to_idx)

    def __contains__(self, logo_id: int) -> bool:
        return self.get_index(logo_id) is not None

    def get_logo_ids(self) -> Iterable[int]:
        """Return an iterable of the external IDs of stored logos."""
        return self.logo_id_to_idx.keys()

    def get_index(self, logo_id: int) -> Optional[int]:
        """Return the index associated with a logo in the ANN index.
        
        :param logo_id: The external ID of the logo
        """
        return self.logo_id_to_idx.get(logo_id)

    def get_embedding(self, logo_id: int) -> Optional[np.ndarray]:
        """Return the embedding of a logo.
        
        :param logo_id: The external ID of the logo
        """
        idx = self.get_index(logo_id)

        if idx is None:
            return None

        if self.hdf5_path.is_file():
            with h5py.File(self.hdf5_path, "r") as f:
                embedding_dset = f["embedding"]
                return embedding_dset[idx]

        return None

    @staticmethod
    def load_logo_id_to_index(hdf5_path: pathlib.Path) -> Dict[int, int]:
        """Read the HDF5 file and generate the logo ID to index mapping."""
        if hdf5_path.is_file():
            with h5py.File(hdf5_path, "r") as f:
                external_id_dset = f["external_id"]
                array = external_id_dset[:]
                non_zero_indexes = np.flatnonzero(array)
                array = array[: non_zero_indexes[-1] + 1]
                return {int(x): i for i, x in enumerate(array)}

        return {}

    def iter_embeddings(self) -> Iterable[Tuple[int, np.ndarray]]:
        """Iterate over HDF5 file and yield (logo_id, embedding) tuples."""
        if not self.hdf5_path.is_file():
            return

        idx_logo_id = sorted(
            ((idx, logo_id) for logo_id, idx in self.logo_id_to_idx.items()),
            key=operator.itemgetter(0),
        )

        with h5py.File(self.hdf5_path, "r") as f:
            embedding_dset = f["embedding"]
            for idx, logo_id in idx_logo_id:
                embedding = embedding_dset[idx]
                yield logo_id, embedding

    def save_embeddings(
        self,
        embeddings: np.ndarray,
        external_ids: np.ndarray,
    ):
        """Add new logo embeddings to the store.
        
        :param embeddings: a numpy array of embeddings, of shape
        (num_logos, embedding_size).
        :param external_ids: a numpy array of external IDs (integer) of shape
        (num_logos, ).
        """
        file_exists = self.hdf5_path.is_file()

        with h5py.File(self.hdf5_path, "a") as f:
            if not file_exists:
                embedding_dset = f.create_dataset(
                    "embedding",
                    (settings.DEFAULT_HDF5_COUNT, embeddings.shape[-1]),
                    dtype="f",
                    chunks=True,
                )
                external_id_dset = f.create_dataset(
                    "external_id",
                    (settings.DEFAULT_HDF5_COUNT,),
                    dtype="i",
                    chunks=True,
                )
            else:
                embedding_dset = f["embedding"]
                external_id_dset = f["external_id"]

            slicing = slice(self.offset, self.offset + len(embeddings))
            embedding_dset[slicing] = embeddings
            external_id_dset[slicing] = external_ids

            for external_id, idx in zip(
                external_ids, range(self.offset, self.offset + len(embeddings))
            ):
                self.logo_id_to_idx[int(external_id)] = idx

            self.offset += len(embeddings)


EMBEDDING_STORE = EmbeddingStore(settings.EMBEDDINGS_HDF5_PATH)


def generate_embeddings(model, images: np.ndarray, device: torch.device) -> np.ndarray:
    images = np.moveaxis(images, -1, 1)  # move channel dim to 1st dim

    with torch.no_grad():
        torch_images = torch.tensor(images, dtype=torch.float32, device=device)
        embeddings = model.extract_features(torch_images).cpu().numpy()

    return np.max(embeddings, (-1, -2))


def crop_image(
    image: Image.Image, bounding_box: Tuple[float, float, float, float]
) -> Image.Image:
    """Crop the image given provided bounding box (y_min, x_min, y_max, x_max),
    with upper-left corner as (0, 0)."""
    y_min, x_min, y_max, x_max = bounding_box
    (left, right, top, bottom) = (
        x_min * image.width,
        x_max * image.width,
        y_min * image.height,
        y_max * image.height,
    )
    return image.crop((left, top, right, bottom))


def add_logos(
    image: Image.Image,
    external_ids: List[int],
    bounding_boxes: List[Tuple[float, float, float, float]],
    device: Optional[torch.device] = None,
) -> int:
    """Compute embeddings for logo extracted from a image and add them to the
    EmbeddingStore.
    
    We only add logos that do not already exist in the EmbeddingStore.

    :param image: The base image
    :param external_ids: The list of external IDs (LogoAnnotation.id)
    :param bounding_box: The bounding boxes of the detected logos. The list must
    be of the same size as `external_ids`
    :param device: The torch device to use to compute the embeddings
    """
    if device is None:
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    model = ModelStore.get(settings.DEFAULT_MODEL, device)
    image_dim = settings.IMAGE_INPUT_DIM[settings.DEFAULT_MODEL]

    selected_external_ids = []
    selected_bounding_boxes = []

    for (bounding_box, external_id) in zip(bounding_boxes, external_ids):
        if external_id in EMBEDDING_STORE:
            continue

        selected_external_ids.append(external_id)
        selected_bounding_boxes.append(bounding_box)

    if not selected_bounding_boxes:
        return 0

    images = np.zeros((len(selected_bounding_boxes), image_dim, image_dim, 3))
    for i, bounding_box in enumerate(selected_bounding_boxes):
        cropped_image = crop_image(image, bounding_box)
        cropped_image = cropped_image.resize((image_dim, image_dim))
        images[i] = np.array(cropped_image)

    embeddings = generate_embeddings(model, images, device)
    EMBEDDING_STORE.save_embeddings(
        embeddings, np.array(selected_external_ids, dtype="i")
    )
    return len(embeddings)


class ModelStore:
    """Simple class to store in memory the embedding models."""
    store: Dict[str, Any] = {}

    @classmethod
    def get(cls, model_name: str, device: torch.device):
        if model_name not in cls.store:
            model = EfficientNet.from_pretrained(model_name)
            model = model.to(device)
            cls.store[model_name] = model

        return cls.store[model_name]