app.py

import gradio as gr
import spaces
import torch
torch.jit.script = lambda f: f
import requests
from PIL import Image
from timm.data import create_transform
from typing import Any


device = "cpu"

# Prepare the model.
import models
model = models.mambaout_femto(pretrained=True).to(device=device) # can change different model name
model.eval()

# Prepare the transform.
transform = create_transform(input_size=224, crop_pct=model.default_cfg['crop_pct'])

# Download human-readable labels for ImageNet.
response = requests.get("https://git.io/JJkYN")
labels = response.text.split("\n")

@spaces.GPU
def predict(inp):
  inp = transform(inp).unsqueeze(0).to(device=device)
  with torch.no_grad():
    prediction = torch.nn.functional.softmax(model(inp)[0], dim=0)
    confidences = {labels[i]: float(prediction[i]) for i in range(1000)}
  return confidences


title="MambaOut: Do We Really Need Mamba for Vision?"
description="Gradio demo for MambaOut model (Femto) proposed by [MambaOut: Do We Really Need Mamba for Vision?](https://arxiv.org/abs/2405.07992). To use it simply upload your image or click on one of the examples to load them. Read more at [arXiv](https://arxiv.org/abs/2405.07992) and [GitHub](https://github.com/yuweihao/MambaOut)."


gr.Interface(title=title,
             description=description,
             fn=predict,
             inputs=gr.Image(type="pil"),
             outputs=gr.Label(num_top_classes=3),
             examples=["images/Kobe_Bryant_2014.jpg", "images/Kobe_coffee.jpg"]).launch()

# Kobo Bryant image credit: https://en.wikipedia.org/wiki/Kobe_Bryant#/media/File:Kobe_Bryant_2014.jpg
# Kobe coffee image credit: https://aroundsaddleworth.co.uk/wp-content/uploads/2020/01/DSC_0177-scaled.jpg