app.py

import os
import json
from typing import List


import torch
import torch.nn.functional as F
import torchvision.transforms as T
from PIL import Image
from torchvision.transforms._transforms_video import NormalizeVideo

import gradio as gr

# Device on which to run the model
# Set to cuda to load on GPU
device = "cpu"
os.system("wget https://huggingface.co/akhaliq/Omnivore/resolve/main/swinB_checkpoint.torch")
# Pick a pretrained model 
model_name = "omnivore_swinB"
model = torch.hub.load('facebookresearch/omnivore:main', "omnivore_swinB", pretrained=False)
new_dict = {}
for key, value in torch.load('/home/user/app/swinB_checkpoint.torch')['trunk'].items():
    new_dict['trunk.' + key] = value
  
for key, value in torch.load('/home/user/app/swinB_checkpoint.torch')['heads'].items():
    new_dict['heads.' + key] = value

model.load_state_dict(new_dict)

# Set to eval mode and move to desired device
model = model.to(device)
model = model.eval()

os.system("wget https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json")

with open("imagenet_class_index.json", "r") as f:
    imagenet_classnames = json.load(f)

# Create an id to label name mapping
imagenet_id_to_classname = {}
for k, v in imagenet_classnames.items():
    imagenet_id_to_classname[k] = v[1] 
    
os.system("wget https://upload.wikimedia.org/wikipedia/commons/thumb/c/c5/13-11-02-olb-by-RalfR-03.jpg/800px-13-11-02-olb-by-RalfR-03.jpg -O library.jpg")

def inference(img):
    image = img
    image_transform = T.Compose(
    [
        T.Resize(224),
        T.CenterCrop(224),
        T.ToTensor(),
        T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ]
    )
    image = image_transform(image)
    
    # The model expects inputs of shape: B x C x T x H x W
    image = image[None, :, None, ...]
    
    prediction = model(image, input_type="image")
    prediction = F.softmax(prediction, dim=1)
    pred_classes = prediction.topk(k=5).indices
    
    pred_class_names = [imagenet_id_to_classname[str(i.item())] for i in pred_classes[0]]
    return "Top 5 predicted labels: %s" % ", ".join(pred_class_names)
    
inputs = gr.inputs.Image(type='pil')
outputs = gr.outputs.Textbox(label="Output")

title = "Omnivore"

description = "Gradio demo for Omnivore: A Single Model for Many Visual Modalities. 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/2201.08377' target='_blank'>Omnivore: A Single Model for Many Visual Modalities</a> | <a href='https://github.com/facebookresearch/omnivore' target='_blank'>Github Repo</a></p>"


gr.Interface(inference, inputs, outputs, title=title, description=description, article=article, examples=[['library.jpg']]).launch(enable_queue=True,cache_examples=True)