app.py

#!/usr/bin/env python
# coding: utf-8

# In[ ]:


import albumentations as A
from albumentations.pytorch.transforms import ToTensorV2
from timm import create_model
import torch
import gradio as gr


# In[ ]:


class TestDataset(torch.utils.data.Dataset):
    def __init__(self,image,transforms = None):
        self.image = [image]
        self.transforms = transforms
    
    def __getitem__(self,idx):
        image = self.image[idx]
        if self.transforms: 
            augmented  = self.transforms(image=image)
            image      = augmented["image"]
        return {'image':image}
    def __len__(self):
        return len(self.image)

def get_test_transform():
    MEAN = [0.5176, 0.4169, 0.3637]
    STD  = [0.3010, 0.2723, 0.2672]
    return A.Compose([
              #A.resize((256,256)),
              A.Normalize(MEAN,STD),
              ToTensorV2(transpose_mask=False,p=1.0)
            ])


# In[ ]:


def predict_image(image):
    test_dataset = TestDataset(image,transforms = get_test_transform())
    test_loader  = torch.utils.data.DataLoader(test_dataset,
                                               batch_size  = 1,
                                               pin_memory  = False,
                                               num_workers = 8,
                                               shuffle     = False)
    # Loading weights
    for data in test_loader:
        for key,value in data.items():
            data[key] = value.to('cpu')
        # Appending Output and Targets:
        output = torch.sigmoid(model(data['image'])).cpu().detach().numpy()
    dict_ = {'Down':float(1-output[0][0]),'Upside':float(output[0][0])}
    return dict_


# In[ ]:


model = create_model('resnet18',pretrained = False,num_classes = 1)
checkpoint = torch.load('model.pt',map_location = 'cpu')
model.load_state_dict(checkpoint,strict = False)


# In[ ]:


title = "Upside-Down Detector"
interpretation='default'
enable_queue=True
gr.Interface(fn=predict_image,inputs=gr.inputs.Image(shape=(256, 256)),outputs=gr.outputs.Label(num_top_classes=2),title=title,interpretation=interpretation).launch()