app.py

# -*- coding: utf-8 -*-
"""
Created on Sat Dec  3 18:31:26 2022

@author: gabri
"""

import numpy as np
import tensorflow as tf
import gradio as gr
from huggingface_hub import from_pretrained_keras
import cv2
import requests
from PIL import Image
import matplotlib.cm as cm
# import matplotlib.pyplot as plt



models_links = {
    'convnext':'https://huggingface.co/gabri14el/grapevine_classification/resolve/main/experimentos/fine-tuning/huge_classifier_20varieties.h5'}

model_weights = {
    }

model_last_convolutional_layer = {'convnext': 'convnext_base_stage_3_block_2_depthwise_conv'}

classes = ['Alveralhao',
 'Arinto do Douro',
 'Cercial',
 'Codega',
 'Codega do Larinho',
 'Donzelinho',
 'Folgasao',
 'Malvasia Fina',
 'Malvasia Preta',
 'Malvasia Rei',
 'Moscatel Galego',
 'Mourisco Tinto',
 'Rabigato',
 'Samarrinho',
 'Sousao',
 'Tinta Amarela',
 'Tinta Barroca',
 'Tinta Roriz',
 'Tinto Cao',
 'Touriga Nacional']
# functions for inference                   
target_size_dimension = 224
n_classes = len(classes)

def define_model(model):
    weights = get_weights(model)
    if model == 'convnext':
        preprocessing_function=tf.keras.applications.convnext.preprocess_input
        model = tf.keras.applications.ConvNeXtBase(
            include_top=False,
            input_shape= (target_size_dimension, target_size_dimension, 3),
            weights='imagenet',
            pooling='avg'
        )
    
    
    x = tf.keras.layers.Dense(512, activation='relu')(model.output)
    x = tf.keras.layers.Dropout(0.20)(x)
    x = tf.keras.layers.Dense(512, activation='relu')(x)
    x = tf.keras.layers.Dropout(0.20)(x)
    output = tf.keras.layers.Dense(n_classes, activation='softmax')(x)
    nmodel = tf.keras.models.Model(model.input, output)
    nmodel.load_weights(weights)
    return preprocessing_function, nmodel

def get_weights(model):
    if not model in model_weights: 
        r = requests.get(models_links[model], allow_redirects=True)
        open(model+'.h5', 'wb').write(r.content)
        model_weights[model] = model+'.h5'
    return model_weights[model]




def get_img_array(img_path, size, expand=True):
    # `img` is a PIL image of size 299x299
    img = tf.keras.preprocessing.image.load_img(img_path, target_size=size)
    # `array` is a float32 Numpy array of shape (299, 299, 3)
    array = tf.keras.preprocessing.image.img_to_array(img)
    # We add a dimension to transform our array into a "batch"
    # of size (1, 299, 299, 3)
    
    if expand:
      array = np.expand_dims(array, axis=0)
    return array


def make_gradcam_heatmap(img_array, grad_model, last_conv_layer_name, pred_index=None, tresh=0.1):
    # First, we create a model that maps the input image to the activations
    # of the last conv layer as well as the output predictions
    #grad_model = tf.keras.models.Model(
        #[model.inputs], [model.get_layer(last_conv_layer_name).output, model.output]
    #)

    # Then, we compute the gradient of the top predicted class for our input image
    # with respect to the activations of the last conv layer
    
    with tf.GradientTape() as tape:
        last_conv_layer_output, preds = grad_model(img_array)
        if pred_index is None:
            pred_index = tf.argmax(preds[0])
        class_channel = preds[:, pred_index]

    # This is the gradient of the output neuron (top predicted or chosen)
    # with regard to the output feature map of the last conv layer
    grads = tape.gradient(class_channel, last_conv_layer_output)

    # This is a vector where each entry is the mean intensity of the gradient
    # over a specific feature map channel
    pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))

    # We multiply each channel in the feature map array
    # by "how important this channel is" with regard to the top predicted class
    # then sum all the channels to obtain the heatmap class activation
    last_conv_layer_output = last_conv_layer_output[0]
    heatmap = last_conv_layer_output @ pooled_grads[..., tf.newaxis]
    heatmap = tf.squeeze(heatmap)

    # For visualization purpose, we will also normalize the heatmap between 0 & 1
    heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)
    heatmap = heatmap.numpy()
    return heatmap

def save_and_display_gradcam(img, heatmap, cam_path="cam.jpg", alpha=0.4):

    # Rescale heatmap to a range 0-255
    heatmap = np.uint8(255 * heatmap)
    im = Image.fromarray(heatmap)
    im = im.resize((img.shape[1], img.shape[0]))
    
    im = np.asarray(im)
    im = np.where(im > 0, 1, im)

    # Use jet colormap to colorize heatmap
    jet = cm.get_cmap("jet")

    # Use RGB values of the colormap
    jet_colors = jet(np.arange(256))[:, :3]
    jet_heatmap = jet_colors[heatmap]

 

    # Create an image with RGB colorized heatmap
    jet_heatmap = tf.keras.preprocessing.image.array_to_img(jet_heatmap)
    jet_heatmap = jet_heatmap.resize((img.shape[1], img.shape[0]))
    jet_heatmap = tf.keras.preprocessing.image.img_to_array(jet_heatmap)

    # Superimpose the heatmap on original image
    superimposed_img = jet_heatmap * alpha + img
    superimposed_img = tf.keras.preprocessing.image.array_to_img(superimposed_img)

    # Save the superimposed image
    #superimposed_img.save(cam_path)

    # Display Grad CAM
    #display(Image(cam_path))
    return superimposed_img, im
        

def infer(model_name, input_image):
    print('#$$$$$$$$$$$$$$$$$$$$$$$$$ IN INFER $$$$$$$$$$$$$$$$$$$$$$$')
    print(model_name, type(input_image))
    preprocess, model = define_model(model_name)
    
    #img = get_img_array(input_image, (target_size_dimension, target_size_dimension))
    img_processed = preprocess(np.expand_dims(input_image, axis=0))
    
    predictions = model.predict(img_processed)
    predictions = np.squeeze(predictions)
    
    
    result = {}
    
    

    
    for i in range(len(classes)):
        result[classes[i]] = float(predictions[i])
    #predictions = np.argmax(predictions) # , axis=2
    #predicted_label = classes[predictions.item()]
    print(input_image.shape)
    model.layers[-1].activation = None
    grad_model = tf.keras.models.Model([model.inputs], [model.get_layer(model_last_convolutional_layer[model_name]).output, model.output])
    
    print(result)
    heatmap = make_gradcam_heatmap(img_processed, grad_model,model_last_convolutional_layer[model_name])
    heat, mask = save_and_display_gradcam(input_image, heatmap)
    
    return result, heat
    
    gr.outputs.Image()
# get the inputs
css = css = ".output-image, .input-image, .image-preview {height: 300px !important}"
inputs = [gr.Radio(["convnext"], label='Choose a model'), gr.inputs.Image(shape=(target_size_dimension, target_size_dimension), label='Select an image')]
# the app outputs two segmented images
output = [gr.outputs.Label(label="Result"), gr.outputs.Image(type="numpy", label="Heatmap (Grad-CAM)")]
# it's good practice to pass examples, description and a title to guide users
examples = [["./content/examples/Frog.jpg"], ["./content/examples/Truck.jpg"]] 
title = "Grapevine image classification"
description = "Upload an image to classify it. The allowed classes are - Alveralhao, Arinto do Douro, Cercial, Codega, Codega do Larinho, Donzelinho, Folgasao, Malvasia Fina, Malvasia Preta, Malvasia Rei, Moscatel Galego, Mourisco Tinto, Rabigato, Samarrinho, Sousao, Tinta Amarela, Tinta Barroca, Tinta Roriz, Tinto Cao, and Touriga Nacional <p><b>Space author: Gabriel Carneiro</b> <br><b> [email protected] </b> </p>"

gr_interface = gr.Interface(infer, inputs, output, allow_flagging=False, analytics_enabled=False, css=css, title=title, description=description).launch(enable_queue=True, debug=False)
#gr_interface.launch()