model.py

import timm
import torch
from torch import nn
import albumentations
from PIL import Image
import numpy as np

augmentations = albumentations.Compose(
    [
        albumentations.Normalize(
            mean=[0.485, 0.456, 0.406],
            std=[0.229, 0.224, 0.225],
            max_pixel_value=255.0,
            always_apply=True
        )
    ]
)

target_map = {
    0: 'Cranberry',
    1: 'Musk melon',
    2: 'Pineapple',
    3: 'Watermelon',
    4: 'Orange',
    5: 'Dragon fruit',
    6: 'Bananas',
    7: 'Blue berries',
    8: 'Jack fruit',
    9: 'Avacados',
}



class ImageModelInfer(nn.Module):
    def __init__(self, model_path, num_classes):
        super().__init__()
        model_path = model_path
        self.model = timm.create_model(model_path, pretrained=False, num_classes=num_classes)
        
    def forward(self, data):
        logits = self.model(data)
        return logits


def prepare_image(image):
    # image = Image.open(image)
    # image = image.convert("RGB")
    # image = image.resize((256, 256), resample=Image.BILINEAR)
    image = np.array(image)
    augmented = augmentations(image=image)
    image = augmented['image']
    image = np.transpose(image, (2, 0, 1)).astype(np.float32)
    return torch.tensor(image, dtype=torch.float)


def predict_fruit_type(img):
    img = prepare_image(img)
    prediction = model(img.unsqueeze(0))[0].detach().numpy()
    class_ = np.argmax(prediction)
    return target_map[class_]


model = ImageModelInfer('vgg16', num_classes=10)
model.load_state_dict(torch.load('best_loss_0.ckpt', map_location=torch.device('cpu'))['state_dict']);