import os

import torch

from PIL import Image

from torchvision import transforms

import gradio as gr

#https://huggingface.co/spaces/yuhe6/final_project/blob/main/Net_Rotate9.pth

#os.system("wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt")

 

#model = torch.hub.load('huawei-noah/ghostnet', 'ghostnet_1x', pretrained=True)

#model = torch.jit.load('https://huggingface.co/spaces/yuhe6/final_project/blob/main/Net_Rotate9.pth').eval().to(device)

 

model = torch.jit.load('Net2_Blur_jit.pt', map_location = torch.device('cpu'))

model.eval()

 

#torch.hub.download_url_to_file('https://huggingface.co/spaces/yuhe6/final_project/blob/main/Net_Rotate9.pth', '/tmp/temporary_file')

#model = torch.hub.load('/tmp', 'temporary_file', pretrained=True)

 

#model.eval()

# Download an example image from the pytorch website

torch.hub.download_url_to_file("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")

 

def inference(input_image):

    preprocess = transforms.Compose([

        transforms.Resize(size = (256, 256)), # Fixed resize from transforms.Resize(256)

        #transforms.CenterCrop(224),

        transforms.ToTensor(),

        #transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),

    ])

 

    # Used print statements to detect shapes between input tensor & batch

    # e.g. input_tensor.shape

    input_tensor = preprocess(input_image)

    input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model

 

    # move the input and model to GPU for speed if available

    if torch.cuda.is_available():

        input_batch = input_batch.to('cuda')

        model.to('cuda')

 

    with torch.no_grad():

        output = model(input_batch) # model(input_tensor) # needed to have batch dimension

 

    # The output has unnormalized scores. To get probabilities, you can run a softmax on it.

    probabilities = torch.nn.functional.softmax(output[0])

 

   # Read the categories

    #with open("dog_cat.txt", "r") as f:

        #categories = [s.strip() for s in f.readlines()]

    #with open("dog_cat.txt", "r") as f:

    categories = ["dog","cat"]

 

    #categories = [s.strip() for s in f.readlines()]

    # Show top categories per image

    top1_prob, top1_catid = torch.topk(probabilities, 1)

    result = {}

    for i in range(top1_prob.size(0)):

        result[categories[top1_catid[i]]] = top1_prob[i].item()

    return result

 

inputs = gr.inputs.Image(type='pil')

outputs = gr.outputs.Label(type="confidences",num_top_classes=1)

 

title = "GHOSTNET"

description = "Gradio demo for GHOSTNET, Efficient networks by generating more features from cheap operations. To use it, simply upload your image, or click one of the examples to load them. Read more at the links below."

article = "<p style='text-align: center'><a href='https://arxiv.org/abs/1911.11907'>GhostNet: More Features from Cheap Operations</a> | <a href='https://github.com/huawei-noah/CV-Backbones'>Github Repo</a></p>"

 

examples = [

            ['dog.jpg']

]

gr.Interface(

  inference, inputs, outputs,

  title = title, description = description,

  article = article, examples = examples,

  analytics_enabled = False).launch(

    #debug = True # Enabled debug mode to see the stacktrace on Google Colab.

  )