app.py

#!/bin/python3

import gradio as gr
import numpy as np
import pandas as pd
import glob, os
import shoe_outlines_lib as sol
import matplotlib.pyplot as plt
import onnxruntime
import cv2

imagenet_stats = ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
imagenet_means = np.array(imagenet_stats[0], dtype=np.float32)[:, None, None]
imagenet_stds  = np.array(imagenet_stats[1], dtype=np.float32)[:, None, None]
sz = (160, 256)

# Load the ONNX model
ort_session = onnxruntime.InferenceSession('shod-model.onnx')


def csv2image_fig(csv_file):
    df = sol.csv2dfs([csv_file])[0]
    fname = df.name
    df = pd.concat([df, df.iloc[[0]]], ignore_index=True) 
    df = sol.norm_by_x(df)
    image = sol.coordsdf2image(df)
    fig = plt.figure(figsize=(2, 4))
    plt.plot(df['x'], df['y'], marker='', linestyle='-', color='b', label='Line')
    plt.fill(df['x'], df['y'], color='blue', alpha=0.2)
    plt.axis('equal')
    plt.axis('off')
    plt.gca().invert_yaxis()
    return image, fig, fname

    
def get_predictions(images, bs=8):
    ''' class 0 is "No shoe", class 1 is "Shoe" '''

    def _softmax(logits):
        exp_logits = np.exp(logits - np.max(logits, axis=-1, keepdims=True))  # Stability trick
        return exp_logits / np.sum(exp_logits, axis=-1, keepdims=True)
    
    if isinstance(images, np.ndarray): images = [images]
    
    images = np.stack([cv2.resize(image, sz) for image in images])
    images = images.transpose(0,3,1,2).astype(np.float32)
    images = (images / 255.0 - imagenet_means) / imagenet_stds

    for b in range(0, len(images), bs):
        ort_inputs = {ort_session.get_inputs()[0].name: images[b:b+bs]}
        preds = ort_session.run(None, ort_inputs)[0]
        all_preds = preds if b==0 else np.concatenate((all_preds, preds))
    confidences = _softmax(all_preds)[:,1] # class 0 is "Bare", class 1 is "Shod"
    
    return confidences


css = """
h1 {
    text-align: center;
    display:block;
    vertical-align: middle;
}
#title-column {
    padding: 0px !important; /* Remove padding from the parent column */
    gap: 0px !important; /* Ensure gap is zero */
}
#title-and-subtitle {
     margin: 0px !important;
     padding: 0px !important;
}
#title-and-subtitle .prose h1 {
    margin: 0px !important;
    padding: 0px !important;
}
#title-and-subtitle .prose p {
    margin: 0px !important;
    padding: 0px !important;
    color: gray !important; /* Added */
    text-align: center !important; /* Added */
    font-style: italic !important; /* Added */
}
.logo {
    max-height: 128px;
    display: inline-block;
    vertical-align: middle;
}
.gradio-container {
    width: 1200px !important; /* Use !important to override defaults if needed */
    margin: 0 auto;
}
"""

with gr.Blocks(css=css) as app:
    with gr.Column():
        with gr.Row():
            gr.Image(
                value="paleostep-logo-cropped-128.png",
                interactive=False,
                show_label=False,
                show_download_button=False,
                show_share_button=False, 
                container=False, 
                show_fullscreen_button=False,
                elem_id="logo",
            )
        with gr.Row():
            with gr.Column(elem_id="title-column"):
                gr.Markdown("""
                    # STEP: Shod Track Estimated Percentage
                    <p style='color: gray; text-align: center; font-style: italic; margin: 0; padding: 0;'>Mysteriously Accurate Rim Curvature INdex</p>
                    """, elem_id="title-and-subtitle")

    #################################################################################
    with gr.Tab('Single outline classification'):
        with gr.Row():
            gr_input = gr.File(file_types=['.csv', '.xlsx', '.json'], file_count="single", label="Upload Outline File")

        with gr.Row():
            gr.Label(value="Upload a .csv/.xlsx/.json file", visible=True, show_label=False)

        with gr.Row():
            gr_plot = gr.Plot(label="Outline Plot", show_label=True, visible=False)

        with gr.Row():
            gr_label = gr.Label(label="Classification", visible=False, show_label=False)
    
    def _classify_image(csv_file):
        try:
            image, fig, fname = csv2image_fig(csv_file)
            if len(image.shape) == 2: image = np.tile(image[...,None],(1,1,3))
            confidence = get_predictions([image]).item()
            classification = "Shoe" if confidence >= 0.5 else "No shoe"
            return (
                classification, {f"Shoe confidence: {100*confidence:.1f}": confidence}, gr.update(visible=True), # gr_label
                fig, gr.update(visible=True, label=fname) # gr_plot
            )
        except Exception as e:
            return str(e), str(e), gr.update(visible=True), None, gr.update(visible=False)
    
    gr_input.upload(
        fn=_classify_image,
        inputs=[gr_input],
        outputs=[gr_label, gr_label, gr_label, gr_plot, gr_plot],
    )

    gr_input.clear(
        fn=lambda: (*([None]*2), *([gr.update(visible=False)]*2)),
        inputs=[],
        outputs=[gr_label, gr_plot, gr_label, gr_plot],
    ) 


    ################################################################################# 
    with gr.Tab('Batch classification'):
        with gr.Row():
            gr_input_batch = gr.File(file_types=['.csv', '.xlsx', '.json'], file_count="multiple", label="Upload Outline File(s)")   
        with gr.Row():
            gr.Label(value="Upload multiple .csv/.xlsx/.json files.", visible=True, show_label=False)
        with gr.Row(visible=True):
            with gr.Column():
                gr_df = gr.Dataframe(label="Outlines", visible=False, show_label=False, row_count=10)
                gr_results_file = gr.File(visible=False)
    
    def _classify_batch(csv_files):
        try:
            for f in glob.glob("classification_results_*.csv"): 
                os.remove(f)

            dfs = sol.csv2dfs(csv_files)
            images = [np.tile(sol.coordsdf2image(df)[...,None],(1,1,3)) for df in dfs]
            confidences = get_predictions(images)

            out = []
            for df, confidence in zip(dfs,confidences):
                images.append(sol.coordsdf2image(df))
                out.append({
                    'Outline file': df.name,
                    'Points': len(df),
                    'Confidence': 100*confidence
                })

            df_out = pd.DataFrame(out)
            timestamp = pd.Timestamp.now().strftime('%Y%m%d_%H%M%S')
            filename = f"classification_results_{timestamp}.csv"
            df_out.to_csv(filename, index=False)

            return df_out.style.format({'Confidence': '{:.1f}%'}), gr.update(visible=True), gr.update(visible=True, value=filename)
        
        except Exception as e:
            return pd.DataFrame({'Error': [str(e)]}), gr.update(visible=True), gr.update(visible=False)

    gr_input_batch.upload(
        fn=_classify_batch,
        inputs=[gr_input_batch],
        outputs=[gr_df, gr_df, gr_results_file],
    )

    gr_input_batch.clear(
        fn=lambda: (None, *([gr.update(visible=False)]*2)),
        inputs=[],
        outputs=[gr_df, gr_df, gr_results_file],
    ) 


app.launch(
    share=False,
    debug=False,
    show_api=False
)