app.py

import gradio as gr
from fastai.vision.all import *
import pathlib
plt = platform.system()
if plt == 'Linux': pathlib.WindowsPath = pathlib.PosixPath

def get_x(r):
    return r['name']


def get_y(r):
    return r['labels'].split(' ')

learner = load_learner('model.pkl')

labels = learner.dls.vocab

def bla(predicted):
    ShirtLength = ('Crop_length', 'Regular_length', 'Long_length', 'ShirtLength_other')
    ShirtNeck = ('Round_neck', 'Tailored_collar_neck', 'Turtle_neck', 'V_neck', 'ShirtNeck_other')
    ShirtSleeveLength = ('Short_sleeve', 'Long_sleeve', 'Sleeveless', 'ShirtSleeveLength_other')
    PatternPlacement = ('No_pattern', 'Pattern')

    shirtlength_idx = [labels.o2i[s] for s in ShirtLength]
    shirtneck_idx = [labels.o2i[s] for s in ShirtNeck]
    shirtsleevelength_idx = [labels.o2i[s] for s in ShirtSleeveLength]
    patternplacement_idx = [labels.o2i[s] for s in PatternPlacement]

    shirtlength_pred = predicted[2][shirtlength_idx]
    shirtneck_pred = predicted[2][shirtneck_idx]
    shirtsleevelength_pred = predicted[2][shirtsleevelength_idx]
    patternplacement_pred = predicted[2][patternplacement_idx]

    val, ind = shirtlength_pred.sort(descending=True)
    #l1 = {ShirtLength[i]: float(shirtlength_pred[i]) for i in ind}
    l1 = {ShirtLength[ind[0]]: float(shirtlength_pred[ind[0]])}
    
    val, ind = shirtneck_pred.sort(descending=True)
    #l2 = {ShirtNeck[i]: float(shirtneck_pred[i]) for i in ind}
    l2 = {ShirtNeck[ind[0]]: float(shirtneck_pred[ind[0]])}
          
    val, ind = shirtsleevelength_pred.sort(descending=True)
    #l3 = {ShirtSleeveLength[i]: float(shirtsleevelength_pred[i]) for i in ind}
    l3 = {ShirtSleeveLength[ind[0]]: float(shirtsleevelength_pred[ind[0]])}

    val, ind = patternplacement_pred.sort(descending=True)
    #l4 = {PatternPlacement[i]: float(patternplacement_pred[i]) for i in ind}
    l4 = {PatternPlacement[ind[0]]: float(patternplacement_pred[ind[0]])}
    
    
    l1.update(l2)
    l1.update(l3)
    l1.update(l4)
    return l1

def predict(img):
    img = PILImage.create(img)
    # pred,pred_idx,probs = learner.predict(img)
    # return {labels[i]: float(probs[i]) for i in range(len(labels))}
    
    pred = learner.predict(img)
    return bla(pred)

title = "Multi-Class Classifier"
description = "Fasion multi-class classifier"
article="<p style='text-align: center'><a href='https://tmabraham.github.io/blog/gradio_hf_spaces_tutorial' target='_blank'>Blog post</a></p>"
examples = ['demo1.jpg', 'demo2.jpg', 'demo3.jpg', 'demo4.jpg', 'demo5.jpg']
interpretation='default'
enable_queue=True

gr.Interface(fn=predict,
             inputs=gr.inputs.Image(shape=(300, 300)),
             outputs=gr.outputs.Label(),
             title=title,
             description=description,
             article=article,
             examples=examples,
             interpretation=interpretation,
             enable_queue=enable_queue).launch()