beerds/server/modules/recognizer/service.py

import asyncio
import io
import os
from dataclasses import dataclass
from datetime import UTC, datetime
from typing import Any, NewType, Protocol
from uuid import uuid4

from dataclasses_ujson.dataclasses_ujson import UJsonMixin  # type: ignore
from PIL import Image

os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import torch
from torchvision import transforms  # type: ignore

from server.modules.attachments.domains.attachments import Attachment
from server.modules.descriptions.repository import ACharactersRepository, Breed
from server.modules.recognizer.repository import ARecognizerRepository, models

TorchModel = NewType("TorchModel", torch.nn.Module)


def load_model(model_path, device="cpu") -> TorchModel:
    model = torch.load(model_path, map_location=device, weights_only=False)
    model.eval()
    return TorchModel(model)


DOG_MODEL = load_model("server/models/dogs_model.pth")
CAT_MODEL = load_model("server/models/cats_model.pth")


class AttachmentService(Protocol):
    async def create(self, file: bytes) -> Attachment:
        pass


@dataclass
class ResultImages(UJsonMixin):
    name: str
    url: list[str]


@dataclass
class RecognizerResult(UJsonMixin):
    results: dict
    images: list
    description: dict[str, list] | None
    uploaded_attach_id: str | None

@dataclass
class SharingBeerds(UJsonMixin):
    alias: str
    name: str
    images: list[ResultImages]

@dataclass
class SharingResult(UJsonMixin):
    beerds: list[SharingBeerds]
    attachment_id: str


class RecognizerService:
    __slots__ = ("_repository", "_attachment_service", "_repository_characters")

    def __init__(
        self,
        repository: ARecognizerRepository,
        attachment_service: AttachmentService,
        repository_characters: ACharactersRepository,
    ):
        self._repository = repository
        self._attachment_service = attachment_service
        self._repository_characters = repository_characters

    async def images_cats(self) -> dict:
        return await self._repository.images_cats()

    async def images_dogs(self) -> dict:
        return await self._repository.images_dogs()

    async def create_result(self, attachment: Attachment, user_id: str, device_id: str, beerd_names: list[str]):
        characters = await self._repository_characters.get_characters()
        await self._repository.create_result_with_beerds(
            models.Results(
                id=str(uuid4()),
                attachment_id=attachment.id,
                user_id=user_id,
                device_id=device_id,
                created_at=datetime.now(UTC),
            ),
            [ch.id for ch in characters if ch.name in beerd_names],
        )

    async def get_results(self, result_id: str) -> SharingResult:
        results = await self._repository.get_results()
        beerds_store: dict[str, Breed] = {b.id: b for b in await self._repository_characters.get_characters()}
        images_dogs = await self._repository.images_dogs()
        for r in results:
            if r.result.id != result_id:
                continue
            beers: list[SharingBeerds] = []
            for beerd in r.beerds:
                beers.append(SharingBeerds(
                    alias=beerds_store[beerd.beerd_id].alias,
                    name=beerds_store[beerd.beerd_id].name,
                    images = [ResultImages(name=beerds_store[beerd.beerd_id].name, url=[f"/static/assets/cat/{beerds_store[beerd.beerd_id].name}/{i}" for i in images_dogs[beerds_store[beerd.beerd_id].name.replace(" ", "_")]])]
                ))
        return SharingResult(beerds=beers, attachment_id=r.result.attachment_id)

    async def predict_dog_image(self, image: bytes, user_id: str, device_id: str | None) -> RecognizerResult:
        if device_id is None:
            device_id = "mobile"
        attachment = await self._attachment_service.create(image)
        predicted_data = self._predict(image, DOG_MODEL)
        results = {}
        images = []
        description: dict[str, list] = {}
        images_dogs = await self._repository.images_dogs()
        for d in predicted_data:
            predicted_idx, probabilities = d
            predicted_label: str = self._repository.labels_dogs()[str(predicted_idx)]
            name = predicted_label.replace("_", " ")
            images.append(
                ResultImages(
                    name=name, url=[f"/static/assets/dog/{predicted_label}/{i}" for i in images_dogs[predicted_label]]
                )
            )
            description.setdefault(name, []).append(f"/dogs-characteristics/{name.replace(' ', '_')}")
            results[probabilities] = name
        asyncio.create_task(self.create_result(attachment, user_id, device_id, [results[key] for key in results]))
        return RecognizerResult(
            results=results, images=images, description=description, uploaded_attach_id=attachment.id
        )

    async def predict_cat_image(self, image: bytes) -> RecognizerResult:
        attachment = await self._attachment_service.create(image)
        predicted_data = self._predict(image, CAT_MODEL)
        results = {}
        images = []
        images_cats = await self._repository.images_cats()
        for d in predicted_data:
            predicted_idx, probabilities = d
            predicted_label: str = self._repository.labels_cats()[str(predicted_idx)]
            name = predicted_label.replace("_", " ")
            images.append(
                ResultImages(
                    name=name, url=[f"/static/assets/cat/{predicted_label}/{i}" for i in images_cats[predicted_label]]
                )
            )
            results[probabilities] = name
        return RecognizerResult(results=results, images=images, description=None, uploaded_attach_id=attachment.id)

    def _predict(self, image: bytes, model, device="cpu") -> list[Any]:
        img_size = (224, 224)
        preprocess = transforms.Compose(
            [
                transforms.Resize(img_size),
                transforms.ToTensor(),
                transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
            ]
        )
        input_tensor = preprocess(Image.open(io.BytesIO(image)))
        input_batch = input_tensor.unsqueeze(0).to(device)  # Добавляем dimension для батча

        with torch.no_grad():
            output = model(input_batch)

        probabilities = torch.nn.functional.softmax(output[0], dim=0)
        k = 5
        topk_probs, predicted_idx = torch.topk(probabilities, k)

        predicted_data = []
        for i in range(k):
            predicted_data.append((predicted_idx[i].item(), float(topk_probs[i].item())))
        return predicted_data