app.py

import numpy as np
import pandas as pd
import gradio as gr
import cv2
from tensorflow import keras as k

# ============================== variables ==============================
image_size = 256
num_classes = 3

in_channel_tool = 3
in_channel_spec = 9
img_rows, img_cols = image_size, image_size

# ============================== paths ==============================
model_class_path = f"Models/minape_base_multi_ts.h5"

# dataset
csv_path = "Dataset/labels_sample.csv"
tool_path = "Dataset/tool"
spec_path = "Dataset/spec"

# ============================== load/prepare dataset ==============================
df = pd.read_csv(csv_path)

df["tool"] = df.id.map(lambda id: f"{tool_path}/{id}.jpg")
df["spec_x"] = df.id.map(lambda id: f"{spec_path}/x/{id}.jpg")
df["spec_y"] = df.id.map(lambda id: f"{spec_path}/y/{id}.jpg")
df["spec_z"] = df.id.map(lambda id: f"{spec_path}/z/{id}.jpg")

# ============================== examples ===========================================
exs = []
for i in range(len(df)):
    row = df.iloc[i,:]
    tool_id = row.id
    image_label = row.image_label 
    tool = row.tool 
    spec_x = row.spec_x 
    spec_y = row.spec_y 
    spec_z = row.spec_z 
    example = [tool_id, image_label, tool, spec_x, spec_y, spec_z]
    exs.append(example)

# ============================== preprocess ===========================================
def process_img(img, img_rows, img_cols, channels):
    """
    Reads the spectogram files from disk and normalizes the pixel values
      @params:
        img - Data of the image
        img_rows - The image height.
        img_cols - The image width.
        as_grey - Read the image as Greyscale or RGB.
        channels - Number of channels.
      @returns:
        The created and compiled model (Model)
    """
    img = cv2.imread(img)
    img = cv2.resize(img, dsize=(img_rows, img_cols), interpolation=cv2.INTER_CUBIC)
    img = np.asarray(img, dtype=np.float32)

    # normalize
    #print(np.max(img))
    img = img / 255.0 #np.max(img)

    # reshape to match Keras expectaions
    img = img.reshape(img_rows, img_cols, channels)

    return img

def process_specs(img_x, img_y, img_z, img_rows, img_cols, channels):
    img_x = cv2.imread(img_x)
    img_y = cv2.imread(img_y)
    img_z = cv2.imread(img_z)

    img_x = cv2.resize(img_x, dsize=(img_rows, img_cols), interpolation=cv2.INTER_CUBIC)
    img_y = cv2.resize(img_y, dsize=(img_rows, img_cols), interpolation=cv2.INTER_CUBIC)
    img_z = cv2.resize(img_z, dsize=(img_rows, img_cols), interpolation=cv2.INTER_CUBIC)
    img = np.concatenate([img_x, img_y, img_z], axis=2)
    img = np.asarray(img, dtype=np.float32)

    # normalize
    img = img / 255.0 #np.max(img)
    return img


# ============================== load model ==============================================
model_class = k.models.load_model(model_class_path, compile=False)
    
# ============================== app =====================================================

def predict(tool_id, label, tool, spec_x, spec_y, spec_z):
    labels = ['sharp', 'used', 'dulled']
    tool = process_img(tool, img_rows, img_cols, in_channel_tool)
    spec = process_specs(spec_x, spec_y, spec_z, img_rows, img_cols, in_channel_spec)


    inputs = [np.array([tool,]), np.array([spec,])]
    y_score = model_class.predict(inputs)
    y_pred = {label:float(score) for label, score in zip(labels, y_score[0])}
    return [
        gr.Label(value=label, label="Actual Label", visible=True), gr.Label(value=y_pred, label="Predicted Label", visible=True), 
      ]

# ============================== blocks ========================================================
title = r"""
<h1 align="center">Minape</h1>
"""
description = r"""
<b>Official 🤗 Gradio demo</b> for <a href='https://github.com/hubtru/Minape' 
target='_blank'><b>Multimodal, Isotropic Neural Architecture with Patch Embedding for Recognition of Device State</b></a>.<br>
"""

with gr.Blocks() as demo:
  gr.Markdown(value=title)
  gr.Markdown(description)
  with gr.Row():
    with gr.Column():
      with gr.Row():
        tool_id = gr.Textbox("T1R2B1", label="Tool")
        label_input = gr.Textbox("Sharp", label="Label")
      with gr.Row():
        tool = gr.Image(label="Tool", type="filepath")
      with gr.Row():
        spec_x = gr.Image(label="Spec_x", type="filepath")
        spec_y = gr.Image(label="Spec_y", type="filepath")
        spec_z = gr.Image(label="Spec_z", type="filepath")
      submit_btn = gr.Button("Submit", variant="primary")
      
    with gr.Column():
      output_labels = [
        gr.Label("Sharp", label="Actual Label"),
        gr.Label("Sharp", label="Predicted Label"),
      ]
  
  examples = gr.Examples(examples=exs, inputs=[tool_id, label_input, tool, spec_x, spec_y, spec_z])
  submit_btn.click(fn=predict, inputs=[tool_id, label_input, tool, spec_x, spec_y, spec_z], outputs=output_labels)
demo.launch()