app.py

import os
import random

import cv2
import gradio as gr
import joblib
import numpy as np
import torch

from models.cnn import Classifier
from models.feed_forward import FeedForwardClassifier

simple_classifier = Classifier()
CNN_PATH = "models/classifier_cnn.pth"
simple_classifier.load_state_dict(torch.load(CNN_PATH))

feed_forward_classifier = FeedForwardClassifier(784)
FF_PATH = "models/classifier.pth"
feed_forward_classifier.load_state_dict(torch.load(FF_PATH))

# Required for the classifier
from sklearn.ensemble import RandomForestClassifier

RF_PATH = "models/fashion_mnist_rf_model.pkl"
rf_clf = joblib.load(RF_PATH)


# Required for the classifier
from sklearn.svm import SVC

SVM_PATH = "models/fashion_mnist_svm_model.pkl"
svm_clf = joblib.load(SVM_PATH)
SVM_PATH_RBF = "models/fashion_mnist_svm_model_rbf.pkl"
svm_clf_rbf = joblib.load(SVM_PATH_RBF)


# Required for the classifier
from sklearn.linear_model import LogisticRegression

LR_PATH = "models/fashion_mnist_lr_model.pkl"
lr_clf = joblib.load(LR_PATH)

# Required for the classifier

from sklearn.neighbors import KNeighborsClassifier

KNN_PATH = "models/fashion_mnist_knn_model.pkl"
knn_clf = joblib.load(KNN_PATH)

# Required for the classifier
from xgboost import XGBClassifier

XGB_PATH = "models/fashion_mnist_xgb_model.pkl"
xgb_clf = joblib.load(XGB_PATH)


LABELS = [
    "T-shirt/top",
    "Trouser",
    "Pullover",
    "Dress",
    "Coat",
    "Sandal",
    "Shirt",
    "Sneaker",
    "Bag",
    "Ankle boot",
]


def classify(img: str):
    # read image
    img = cv2.imread(img, cv2.IMREAD_GRAYSCALE)
    img = cv2.resize(img, (28, 28))
    img = img / 255.0
    img = np.array(img).reshape(-1, 1, 28, 28)
    img = torch.from_numpy(img).float()

    # CNN Classifier
    cnn_output = simple_classifier(img)
    cnn_output = torch.nn.functional.softmax(cnn_output, dim=1)
    cnn_output_pred = torch.argmax(cnn_output)
    cnn_output_label = LABELS[cnn_output_pred.item()]
    cnn_confidence = round(torch.max(cnn_output).item() * 100, 2)
    cnn_output_str = (
        f"{'CNN:':<35} {cnn_output_label:<15} with {cnn_confidence:.2f}% confidence"
    )

    # Feed Forward Classifier
    feed_forward_output = feed_forward_classifier(img)
    feed_forward_output = torch.nn.functional.softmax(feed_forward_output, dim=1)
    feed_forward_output_pred = torch.argmax(feed_forward_output)
    feed_forward_output_label = LABELS[feed_forward_output_pred.item()]
    feed_forward_confidence = round(torch.max(feed_forward_output).item() * 100, 2)
    feed_forward_output_str = f"{'Feed Forward:':<35} {feed_forward_output_label:<15} with {feed_forward_confidence:.2f}% confidence"

    # XGBoost Classifier
    xgb_output = xgb_clf.predict(img.reshape(1, -1))
    xgb_output_label = LABELS[xgb_output[0]]
    xgb_confidence = round(
        float(np.max(xgb_clf.predict_proba(img.reshape(1, -1))[0])) * 100, 2
    )
    xgb_output_str = (
        f"{'XGBoost:':<35} {xgb_output_label:<15} with {xgb_confidence:.2f}% confidence"
    )

    # Random Forest Classifier
    rf_output = rf_clf.predict(img.reshape(1, -1))
    rf_output_label = LABELS[rf_output[0]]
    rf_output_str = f"{'Random Forest:':<35} {rf_output_label:<15}"

    # SVM with Linear Kernel Classifier
    svm_output = svm_clf.predict(img.reshape(1, -1))
    svm_output_label = LABELS[svm_output[0]]
    svm_output_str = f"{'SVM with Linear kernel:':<35} {svm_output_label:<15}"

    # SVM with RBF Kernel Classifier
    svm_output_rbf = svm_clf_rbf.predict(img.reshape(1, -1))
    svm_output_label_rbf = LABELS[svm_output_rbf[0]]
    svm_output_str_rbf = f"{'SVM with RBF kernel:':<35} {svm_output_label_rbf:<15}"

    # Logistic Regression Classifier
    lr_output = lr_clf.predict(img.reshape(1, -1))
    lr_output_label = LABELS[lr_output[0]]
    lr_output_str = f"{'Logistic Regression:':<35} {lr_output_label:<15}"

    # KNN Classifier
    knn_output = knn_clf.predict(img.reshape(1, -1))
    knn_output_label = LABELS[knn_output[0]]
    knn_output_str = f"{'KNN:':<35} {knn_output_label:<15}"

    # Combine output
    output = (
        cnn_output_str
        + "\n"
        + feed_forward_output_str
        + "\n"
        + xgb_output_str
        + "\n"
        + rf_output_str
        + "\n"
        + svm_output_str
        + "\n"
        + svm_output_str_rbf
        + "\n"
        + lr_output_str
        + "\n"
        + knn_output_str
    )
    return output


folder = "./images"
examples = []
for filename in os.listdir(folder):
    img_path = os.path.join(folder, filename)
    examples.append([img_path])

random.shuffle(examples)


iface = gr.Interface(
    fn=classify,
    title="Fashion MNIST Classifier - TAECAC @ FEUP",
    description="Simple Proof of Concept.",
    inputs=gr.Image(label="Image", type="filepath"),
    outputs=gr.Textbox(label="Classification output"),
    examples=examples,
    examples_per_page=100,
    theme=gr.themes.Soft(
        primary_hue=gr.themes.colors.indigo,
        secondary_hue=gr.themes.colors.gray,
        neutral_hue=gr.themes.colors.slate,
        font=["avenir"],
    ),
)

iface.launch()