app.py

import sys
import os
import requests
import torch
from PIL import Image
from torchvision import transforms
import gradio as gr


#  timm==0.4.5  # 0.3.2 does not work in Colab

os.system("wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt")
os.system("git clone https://github.com/facebookresearch/mae.git")
sys.path.append('./mae')

import models_mae
import models_vit



def prepare_model(chkpt_dir, arch='vit_large_patch14'):
    # build model
    model = getattr(models_vit, arch)(global_pool=True)
    # load model
    checkpoint = torch.load(chkpt_dir, map_location='cpu')
    msg = model.load_state_dict(checkpoint['model'], strict=True)
    print(msg)
    return model



def inference(input_image):
    preprocess = transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    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)
    # The output has unnormalized scores. To get probabilities, you can run a softmax on it.
    probabilities = torch.nn.functional.softmax(output[0], dim=0)

    # Read the categories
    with open("imagenet_classes.txt", "r") as f:
        categories = [s.strip() for s in f.readlines()]
    # Show top categories per image
    top5_prob, top5_catid = torch.topk(probabilities, 5)
    result = {}
    for i in range(top5_prob.size(0)):
        result[categories[top5_catid[i]]] = top5_prob[i].item()
    return result
    


os.system("wget -nc https://dl.fbaipublicfiles.com/mae/finetune/mae_finetuned_vit_large.pth")
chkpt_dir = 'mae_finetuned_vit_large.pth'
model = prepare_model(chkpt_dir, 'vit_large_patch16')


# Download an example image from the pytorch website
torch.hub.download_url_to_file("https://estaticos.megainteresting.com/media/cache/1140x_thumb/uploads/images/gallery/5e7c585f5cafe8134048af67/gato-persa-gris_0.jpg", "persian_cat.jpg")
torch.hub.download_url_to_file("https://user-images.githubusercontent.com/11435359/147738734-196fd92f-9260-48d5-ba7e-bf103d29364d.jpg", "fox.jpg")
torch.hub.download_url_to_file("https://user-images.githubusercontent.com/11435359/147743081-0428eecf-89e5-4e07-8da5-a30fd73cc0ba.jpg", "cucumber.jpg")

inputs = gr.inputs.Image(type='pil')
outputs = gr.outputs.Label(type="confidences",num_top_classes=5)

title = "MAE"
description = "Gradio demo for Masked Autoencoders (MAE) ImageNet classification (large-patch16). To use it, simply upload your image, or click on the examples to load them. Read more at the links below."

article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2111.06377' target='_blank'>Masked Autoencoders Are Scalable Vision Learners</a> | <a href='https://github.com/facebookresearch/mae' target='_blank'>Github Repo</a></p>"

examples = [
            ['persian_cat.jpg'],
            ['fox.jpg'],
            ['cucumber.jpg']            
]

gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=examples, analytics_enabled=False).launch()