app.py

import os
import json

import torch
import torch.nn.functional as F
import torchvision.transforms as T
from uniformer import uniformer_small
from imagenet_class_index import imagenet_classnames

import gradio as gr

# Device on which to run the model
# Set to cuda to load on GPU
device = "cpu"
os.system("wget https://cdn-lfs.huggingface.co/Andy1621/uniformer/fd192c31f8bd77670de8f171111bd51f56fd87e6aea45043ab2edc181e1fa775")
# Pick a pretrained model 
model = uniformer_small()
state_dict = torch.load('fd192c31f8bd77670de8f171111bd51f56fd87e6aea45043ab2edc181e1fa775', map_location='cpu')
model.load_state_dict(state_dict['model'])

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

# 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.unsqueeze(0)
    
    prediction = model(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]]
    # pred_class_probs = [prediction[0][i.item()].item() * 100 for i in pred_classes[0]]
    # res = "Top 5 predicted labels:\n"
    # for name, prob in zip(pred_class_names, pred_class_probs):
    #     res += f"[{prob:2.2f}%]\t{name}\n"

    return {imagenet_id_to_classname[str(i)]: float(prediction[0][i]) for i in range(1000)}
    
inputs = gr.inputs.Image(type='pil')
# outputs = gr.outputs.Textbox(label="Output")
label = gr.outputs.Label(num_top_classes=5)

title = "UniFormer-S"

description = "Gradio demo for UniFormer: 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/2201.09450' target='_blank'>UniFormer: Unifying Convolution and Self-attention for Visual Recognition</a> | <a href='https://github.com/Sense-X/UniFormer' target='_blank'>Github Repo</a></p>"


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