flow / app.py
jon-fernandes's picture
Updated from colab
e660ff3
raw
history blame
813 Bytes
from transformers import AutoModelForImageClassification, AutoFeatureExtractor
import gradio as gr
import torch
huggingface_username = 'i-am-holmes'
model_name = 'vit-base-patch16-224-finetuned-flower'
def classify_image(image):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = AutoModelForImageClassification.from_pretrained(f'{huggingface_username}/{model_name}').to(device)
feature_extractor = AutoFeatureExtractor.from_pretrained(f'{huggingface_username}/{model_name}')
inp = feature_extractor(image, return_tensors='pt').to(device)
outp = model(**inp)
pred = torch.argmax(outp.logits, dim=1).item()
return model.config.id2label[pred]
interface = gr.Interface(fn=classify_image, inputs=gr.Image(shape=(224, 224)), outputs="text").launch(debug=True, share=True)