# /*--------------------------------------------------------------------------------------------- # * Copyright (c) 2024 STMicroelectronics. # * All rights reserved. # * # * This software is licensed under terms that can be found in the LICENSE file in # * the root directory of this software component. # * If no LICENSE file comes with this software, it is provided AS-IS. # *--------------------------------------------------------------------------------------------*/ import os import re import uuid import time import shutil import zipfile import threading import subprocess import select from datetime import datetime from concurrent.futures import ThreadPoolExecutor import dash from dash import dcc, html import dash_daq as daq from dash.dependencies import Input, Output, State, ALL import dash_bootstrap_components as dbc from dash.exceptions import PreventUpdate import dash_daq as daq from flask import Flask, render_template, request, send_file, jsonify, abort import plotly.graph_objects as go import plotly.colors as pc import yaml import ruamel.yaml import pandas as pd import logging import base64 logging.basicConfig(level=logging.DEBUG) logger = logging.getLogger(__name__) server = Flask(__name__) server.secret_key = os.urandom(24) @server.route('/') def welcome_page(): """ Handles the welcome page route. This function extracts the username from the request host, determines if the duplicate mode should be enabled, and renders the welcome page template with the duplicate mode state. Returns: str: The rendered 'index.html' template with the duplicate_mode parameter. """ host = request.host print("host:", host) usr_match = re.match(r'^(.*?)\-stm32', host) print("usr_match:", usr_match) if usr_match: hf_user = usr_match.group(1) else: hf_user = "modelzoo_user" if hf_user == "stmicroelectronics": duplicate_mode = True else: duplicate_mode = False print("hf_user:", hf_user) print("duplicate_mode:", duplicate_mode) return render_template('index.html', duplicate_mode=duplicate_mode) external_stylesheets = [dbc.themes.LITERA] app = dash.Dash(__name__, server=server,external_stylesheets=external_stylesheets, url_base_pathname='/dash_app/', suppress_callback_exceptions=True) local_yamls = { 'image_classification': 'stm32ai-modelzoo-services/image_classification/src/user_config.yaml', 'human_activity_recognition': 'stm32ai-modelzoo-services/human_activity_recognition/src/user_config.yaml', 'hand_posture': 'stm32ai-modelzoo-services/hand_posture/src/user_config.yaml', 'object_detection': 'stm32ai-modelzoo-services/object_detection/src/user_config.yaml', 'audio_event_detection': 'stm32ai-modelzoo-services/audio_event_detection/src/user_config.yaml', 'pose_estimation': 'stm32ai-modelzoo-services/pose_estimation/src/user_config.yaml', 'semantic_segmentation': 'stm32ai-modelzoo-services/semantic_segmentation/src/user_config.yaml' } def banner(): return html.Div( id="banner", className="top-bar", style={ "display": "flex", "align-items": "center", "justify-content": "space-between", "position": "fixed", "top": "0", "left": "0", "width": "100%", "z-index": "1000", "background": "linear-gradient(to right, #03234b, #054080)", "box-shadow": "0px 4px 8px rgba(0, 0, 0, 0.2)", "border-radius": "0 0 10px 10px" }, children=[ html.A( id="learn-more-button", children=[ html.Img( src=app.get_asset_url("github-mark-white.png"), style={"width": "22px", "height": "22px", "margin-right": "8px"} ), html.Span("stm32ai-modelzoo", style={"font-weight": "bold"}) ], href="https://github.com/STMicroelectronics/stm32ai-modelzoo-services", target="_blank", style={ "display": "flex", "align-items": "center", "color": "#ffffff", "text-decoration": "none", "font-size": "16px", "font-family": "Arial, sans-serif", "transition": "color 0.3s ease" } ), html.Div( [ dbc.Button( html.Img( src=app.get_asset_url("logs.jpg"), style={"width": "22px", "height": "22px","margin-right":"10px"} ), id="toggle-log", n_clicks=0, className="", style={ "background": "none", "border": "none", "padding": "0", "margin-right": "12px", "cursor": "pointer" } ), html.A( html.H5( "ST Edge AI Developer Cloud", style={ "margin": "0", "color": "#ffffff", "font-size": "16px", "font-weight": "bold", "font-family": "Arial, sans-serif", "transition": "color 0.3s ease" } ), href="https://stm32ai-cs.st.com/home", target="_blank", style={ "display": "flex", "align-items": "center", "text-decoration": "none" } ) ], style={"display": "flex", "align-items": "center"} ) ] ) def create_dashboard_layout(): """ Creates the layout for the application: STM32ModelZoo dashboard. This function defines the structure and components of the dashboard, including the banner, model selection dropdown, YAML update options, credentials input, output display, training metrics graphs, and download button. Returns: dbc.Container: A Dash Bootstrap Component container with the dashboard layout. """ return html.Div([ banner(), dbc.Container([ dcc.Location(id='url', refresh=False), dbc.Row(dbc.Col(html.H3("STM32 Model zoo Dashboard", style={'color': '#03234b', 'text-align': 'center', "margin-top": "80px", "font-family": "Arial, sans-serif"}), className="mb-4")), dbc.Row([ dbc.Col( html.H5("Use case selection", style={'color': '#03234b', 'margin-bottom': '10px'}), width=12 ) ], id="use-case-section", style={"display": "none"}), dbc.Row(dbc.Col(dcc.Dropdown( id='selected-model', options=[ {'label': 'Image Classification (IC)', 'value': 'image_classification'}, {'label': 'Human Activity Recognition (HAR)', 'value': 'human_activity_recognition'}, {'label': 'Hand Posture', 'value': 'hand_posture'}, {'label': 'Audio Event Detection(AED)', 'value': 'audio_event_detection'}, {'label': 'Object Detection', 'value': 'object_detection'}, {'label': 'Pose estimation', 'value': 'pose_estimation'}, {'label': 'Semantic Segmentation', 'value': 'semantic_segmentation'}, ], placeholder="Please select your use case", className="mb-4" ))), dbc.Row( dbc.Col( html.Div( id='toggle-yaml', children=[ dbc.Button("How to update User Config ", id="open-offcanvas", n_clicks=0), dbc.Offcanvas( html.Div( [ html.P([ html.Strong("Configure Dataset section:"), html.Br(), "- Dataset path: ../datasets/your_use_case/name_of_dataset or datasets/your_prepared_dataset.", html.Br(), html.Br(), "- For more details, refer to the ", html.A("README", href="https://huggingface.co./spaces/STMicroelectronics/stm32-modelzoo-app/blob/main/datasets/README.md", target="_blank", style={'color': '#007bff', 'text-decoration': 'underline'}), ".", html.Br(), html.Br(), "- If you need to upload your model for evaluation, benchmarking or quantizig:", html.Br(), "- Update model path under General section: models/your_model_name", html.Br(), "- For more details, refer to the ", html.A("README", href="https://huggingface.co./spaces/STMicroelectronics/stm32-modelzoo-app/blob/main/models/README.md", target="_blank", style={'color': '#007bff', 'text-decoration': 'underline'}), "." ], style={'font-family': 'Arial, sans-serif', 'color': '#03234b', 'fontSize': '18px'}) ] ), id="offcanvas", is_open=False, title="📚 Help", placement="end", ), dcc.RadioItems( id='modify-yaml-choice', labelStyle={'display': 'inline-block', 'margin-right': '10px'}, className="mb-4", ), dcc.Upload( id='load-yaml-file', children=html.Button('Upload YAML File'), style={'display': 'none'} ), html.Div(id='load-state', style={'margin-top': '10px'}), html.Div(id='yaml-layout', style={'display': 'none'}) ], style={'font-family': 'Arial, sans-serif'} ) ) ), dbc.Row([ dbc.Col([ html.P("Enter your ST Edge AI Developer Cloud credentials:", style={'color': '03234b', 'fontSize': '15px', 'fontWeight': 'bold'}, className="credentials-text"), dcc.Input(id='devcloud-username-input', type='text', placeholder='Enter username', className="input-field mb-2"), dcc.Input(id='devcloud-password-input', type='password', placeholder='Enter password', className="input-field mb-4") ], width=6), dbc.Col([ dbc.Button('Launch training', id='process-button', color="#ceecf9", className="start-button mb-4", style={'display': 'none', 'box-shadow': '0px 4px 6px rgba(0, 0, 0, 0.1)'}) ], className="credentials-col") ], id='credentials-section', style={ 'display': 'none', 'justify-content': 'center', 'align-items': 'center', 'height': '100vh', }, className="credentials-section mb-4"), dbc.Row([ dbc.Col( html.H5("Results visualization", style={'color': '#03234b', 'margin-bottom': '10px'}), width=12 ) ], id="results-section", style={"display": "none"}), dbc.Row([ dbc.Col(dbc.Card([ dbc.CardHeader("Metrics", style={'background-color': '#03234b', 'color': 'white'}), dbc.CardBody( dcc.Graph(id='acc-visualization', style={'height': '100%', 'width': '100%'}), style={'height': '400px', 'display': 'flex', 'justify-content': 'center', 'align-items': 'center'} ) ]), width=6, style={'padding': '10px'}), dbc.Col(dbc.Card([ dbc.CardHeader("Metrics", style={'background-color': '#03234b', 'color': 'white'}), dbc.CardBody( dcc.Graph(id='loss-visualization', style={'height': '100%', 'width': '100%'}), style={'height': '400px', 'display': 'flex', 'justify-content': 'center', 'align-items': 'center'} ) ]), width=6, style={'padding': '10px'}) ], style={'margin-bottom': '30px'}), dbc.Row([ dbc.Col(dbc.Card([ dbc.CardHeader("Memory Usage", style={'background-color': '#8191a5', 'color': 'white', 'font-size': '20px'}), dbc.CardBody(dcc.Graph(id='memory-bar')) ]), width=4), dbc.Col(dbc.Card([ dbc.CardHeader("Inference Time", style={'background-color': '#8191a5', 'color': 'white', 'font-size': '20px'}), dbc.CardBody(dcc.Graph(id='inference-time')) ]), width=4), ],justify="center"), dbc.Row([ html.Div(id='metric-graphs-container', style={ 'margin-bottom': '30px' }) ]), dcc.Interval(id='interval-widget', interval=1000, n_intervals=0), dcc.Download(id="download-resource"), dbc.Row( dbc.Col( dbc.Button('Download outputs', id='download-action', className="mb-4", style={ 'background-color': '#ffd200', 'color': '#ffffff', 'font-size': '14px', 'padding': '10px 10px', 'border-radius': '5px', 'box-shadow': '0px 4px 6px rgba(0, 0, 0, 0.1)', 'margin-top': '20px' }), style={ 'display': 'flex', 'justify-content': 'center', 'alignItems': 'center', } ) ), dbc.Row([ dbc.Col(dbc.Card([ dbc.CardHeader("Confusion Matrix", style={'background-color': '#8191a5', 'color': 'white', 'font-size': '20px'}), dbc.CardBody( html.Div( html.Img(id='confusion-matrix-img', style={'max-width': '100%', 'height': 'auto'}), style={'display': 'flex', 'justify-content': 'center', 'align-items': 'center', 'height': '100%'} ) ) ]), width=12) ], justify="center") ], fluid=True), dbc.Offcanvas( html.Div(id='log-reader', style={'whiteSpace': 'pre-wrap', 'padding': '15px', 'height': '200px', 'overflow': 'auto'}), id="log-offcanvas", is_open=False, placement="bottom", style={'height': '200px', 'background-color': '#343a40', 'color': 'white', 'resize': 'vertical', 'overflow': 'auto'} ), ]) def read_configs(selected_model): """ Loads a YAML file based on the selected model by the user. Args: selected_model (str): The key to select the appropriate YAML file path. Returns: dict: The loaded YAML data. """ if not selected_model: raise ValueError("No model selected. Please select a valid model.") if selected_model not in local_yamls: raise ValueError(f"Model '{selected_model}' not found in local_yamls") yaml_path = local_yamls[selected_model] try: with open(yaml_path, 'r') as file: return yaml.safe_load(file) except Exception as e: raise ValueError(f"Error reading YAML file at {yaml_path}: {e}") def build_yaml_form(yaml_content, parent_key=''): """ Recursively builds a form based on the provided YAML content. Parameters: - yaml_content (dict): The YAML content to build the form from. - parent_key (str): The parent key to maintain the hierarchy of nested keys. Default is an empty string. Returns: - list: A list of Dash Bootstrap Components (dbc) AccordionItems representing the form fields. """ hidden_sections = {'tools', 'deployment', 'mlflow', 'hydra'} accordion_items = [] for key, value in yaml_content.items(): if key in hidden_sections and parent_key == '': continue full_key = f"{parent_key}.{key}" if parent_key else key if isinstance(value, dict): if full_key == "dataset": section_title = html.Span([ "Dataset ", html.Span("*", style={"color": "red", "fontWeight": "bold"}), html.Span(" (Set dataset path)", style={"fontSize": "0.85rem", "color": "#dc3545", "marginLeft": "5px"}) ]) else: section_title = key.capitalize() nested_accordion = build_yaml_form(value, full_key) accordion_items.append( dbc.AccordionItem( nested_accordion, title=section_title ) ) else: field = [html.Label(key, style={"font-weight": "bold", "margin-bottom": "5px"})] if isinstance(value, bool): field.append( dcc.Checklist( id={'type': 'yaml-setting', 'index': full_key}, options=[{'label': '', 'value': True}], value=[True] if value else [], style={"padding": "10px", "border": "1px solid #ddd", "margin-bottom": "10px"} ) ) elif isinstance(value, list): field.append( dcc.Dropdown( id={'type': 'yaml-setting', 'index': full_key}, options=[{'label': str(v), 'value': v} for v in value], value=value, multi=True, style={"padding": "10px", "border": "1px solid #ddd", "margin-bottom": "10px"} ) ) else: input_style = { "padding": "10px", "border": "1px solid #ddd", "margin-bottom": "10px", "width": "100%" } helper = None if full_key == "dataset.training_path": input_style.update({ "border": "2px solid #ffc107", "backgroundColor": "#fff8e1" }) helper = html.Div( "⚠️ Please update dataset path.", style={ "color": "#856404", "fontSize": "0.85rem", "marginTop": "-8px", "marginBottom": "10px" } ) field.append( dcc.Input( id={'type': 'yaml-setting', 'index': full_key}, value=value, type='text', style=input_style ) ) if helper: field.append(helper) accordion_items.append( dbc.AccordionItem( field, title=key.capitalize() ) ) return accordion_items def create_yaml(yaml_content): """ Creates a YAML form using Dash Bootstrap Components (dbc) and Dash HTML Components (html). Parameters: yaml_content (dict): The content of the YAML file to be used for building the form. Returns: dbc.Form: A Dash form component containing an accordion with the YAML content and a submit button. """ accordion_items = build_yaml_form(yaml_content) accordion = dbc.Accordion( accordion_items, start_collapsed=True ) return dbc.Form([ accordion, html.Div( dbc.Button( 'Submit', id='apply-button', style={ 'background-color': '#FFD200', 'color': '#03234b', 'font-size': '14px', 'padding': '10px 10px 10px 10px', 'border-radius': '5px', 'margin-top': '15px', 'border': '2px solid #FFD200', 'box-shadow': '0px 4px 6px rgba(0, 0, 0, 0.1)', } ), style={ 'display': 'flex', 'justify-content': 'center', 'margin-top': '15px', } ), html.Div( id='submission-outcome', style={ 'marginTop': '10px', 'textAlign': 'center', 'fontStyle': 'italic', 'color': '#03234b', 'font-size': '14px' } ) ]) def process_form_configs(form_configs): """ Extracts and processes form data to update YAML content. This function processes the form data, converting values to appropriate types and updating the YAML content accordingly. Args: form_configs (dict): The form data to be processed. Returns: dict: The updated YAML content with processed form data. """ updated_yaml = {} for key, value in form_configs.items(): if value is not None: if isinstance(value, list) and len(value) == 1: value = value[0] if isinstance(value, str): try: if '.' in value: value = float(value) else: value = int(value) except ValueError: pass updated_yaml[key] = value return updated_yaml def create_archive(archive_path, directory_to_compress): """ Creates a ZIP archive of a specified directory. Parameters: archive_path (str): The path where the ZIP archive will be created. directory_to_compress (str): The directory whose contents will be compressed into the ZIP archive. Returns: None """ def add_file_to_zip(zipf, file_path, arcname): """ Adds a file to the ZIP archive. Parameters: zipf (zipfile.ZipFile): The ZIP file object. file_path (str): The path of the file to add to the ZIP archive. arcname (str): The archive name for the file within the ZIP archive. Returns: None """ zipf.write(file_path, arcname=arcname) with zipfile.ZipFile(archive_path, 'w', compression=zipfile.ZIP_DEFLATED) as zipf: with ThreadPoolExecutor() as executor: for root_dir, sub_dirs, files in os.walk(directory_to_compress): for file_name in files: file_path = os.path.join(root_dir, file_name) if os.path.abspath(file_path) != os.path.abspath(archive_path): arcname = os.path.relpath(file_path, directory_to_compress) executor.submit(add_file_to_zip, zipf, file_path, arcname) app.layout = create_dashboard_layout logs = [] lock = threading.Lock() new_training = False def fill_logs(message): """ Appends a message to the logs list in a thread-safe manner and returns the formatted logs. Parameters: message (str): The message to be appended to the logs. Returns: html.Pre: The formatted logs as HTML content. """ with lock: logs.append(message) filtered_logs = filter_logs("\n".join(logs)) formatted_logs = format_logs(filtered_logs) return html.Pre(formatted_logs, style={'whiteSpace': 'pre-wrap', 'wordBreak': 'break-all'}) def filter_logs(logs): important_lines = [] for line in logs.split('\n'): if '[INFO]' in line or 'Epoch' in line or 'Total params' in line or 'Trainable params' in line or 'Non-trainable params' in line: important_lines.append(line) elif 'Segments built' in line: important_lines.append(line) return '\n'.join(important_lines) def format_logs(logs): formatted_logs = logs.replace('[INFO]', '\n[INFO]').replace('Epoch', '\nEpoch').replace('Segments built', '\nSegments built') return formatted_logs def extract_metrics(logs): metrics = { 'float': {}, 'quantized': {}, 'oks': {}, } float_match = re.search(r"Accuracy of float model(?: on validation_set)?\s*=\s*([\d.]+)\s*%", logs) if float_match: metrics['float']['accuracy'] = float(float_match.group(1)) quant_match = re.search(r"Accuracy of quantized model(?: on validation_set)?\s*=\s*([\d.]+)\s*%", logs) if quant_match: metrics['quantized']['accuracy'] = float(quant_match.group(1)) precision_match = re.search(r"Mean precision:\s*([\d.]+)", logs) if precision_match: metrics['oks']['precision'] = float(precision_match.group(1)) recall_match = re.search(r"Mean recall:\s*([\d.]+)", logs) if recall_match: metrics['oks']['recall'] = float(recall_match.group(1)) ap_match = re.search(r"Mean AP $mAP$:\s*([\d.]+)", logs) if ap_match: metrics['oks']['mean_ap'] = float(ap_match.group(1)) oks_match = re.search(r"The mean OKS is :\s*([\d.]+)", logs) if oks_match: metrics['oks']['mean_oks'] = float(oks_match.group(1)) iou_match = re.search(r"Average IoU of float model \(all classes\) on validation_set\s*=\s*([\d.]+)\s*%", logs) if iou_match: metrics['float']['average_iou'] = float(iou_match.group(1)) return metrics def _parse_inference_memory(logs): metrics = {} """ patterns = { "ram": r"Total RAM\s*:\s*([\d.]+)\s*\(KiB\)", "flash": r"Total Flash\s*:\s*([\d.]+)\s*\(KiB\)", "inference_time": r"Inference Time\s*:\s*([\d.]+)\s*\(ms\)" } """ patterns = { "ram": r"Total RAM\s*:\s*([\d.]+)\s*\(KiB\)", "flash": r"Total Flash\s*:\s*([\d.]+)\s*\(KiB\)", "inference_time": r"Inference Time\s*:\s*([\d.]+)\s*\(ms\)" } for key, pattern in patterns.items(): matches = re.findall(pattern, logs) if matches: metrics[key] = float(matches[-1]) return metrics def create_accuracy_gauge(accuracy): return go.Figure(go.Indicator( mode="gauge+number", value=accuracy, title={'text': "Accuracy (%)"}, gauge={'axis':{'range': [0, 100]}, 'bar':{'color':"#49B170"}}, )) def create_iou_gauge(iou_value): fig = go.Figure(go.Indicator( mode="gauge+number", value=iou_value, title={'text': "IoU (%)"}, gauge={'axis': {'range': [0, 100]}, 'bar': {'color': "#49B170"}} )) return fig def latest_confusion_matrix(outputs_folder, recent_directory): cf_path = os.path.join(outputs_folder, recent_directory) for filename in os.listdir(cf_path): if "confusion_matrix" in filename and filename.endswith(".png"): image_path = os.path.join(cf_path, filename) with open(image_path, "rb") as f: encoded = base64.b64encode(f.read()).decode() return f"data:image/png;base64,{encoded}" return None def run_script(script, devcloud_username, devcloud_password): """ Executes a given script with the provided ST Developer Cloud credentials and logs the output. Parameters: - script (str): The path to the script to be executed. - devcloud_username (str): Username for ST Developer Cloud. - devcloud_password (str): Password for ST Developer Cloud. Returns: - None """ global logs with lock: logs = [] isolated_env = os.environ.copy() isolated_env['stmai_username'] = devcloud_username isolated_env['stmai_password'] = devcloud_password isolated_env['STATS_TYPE'] = 'HuggingFace_devcloud' execution = subprocess.Popen(['python3', script], env=isolated_env, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True) while True: file_descriptors = [execution.stdout.fileno(), execution.stderr.fileno()] selected_descriptors = select.select(file_descriptors, [], []) for descriptor in selected_descriptors[0]: if descriptor == execution.stdout.fileno(): out = execution.stdout.readline() if out: fill_logs(out) if out == '' and execution.poll() is not None: return if descriptor == execution.stderr.fileno(): error = execution.stderr.readline() if error: fill_logs(error) def execute_async(script, devcloud_username, devcloud_password): """ Executes a Python script asynchronously in a separate thread. Parameters: script (str): The path to the Python script to be executed. devcloud_username (str): The username for the DevCloud environment. devcloud_password (str): The password for the DevCloud environment. Returns: None """ thread = threading.Thread(target=run_script, args=(script, devcloud_username, devcloud_password)) thread.start() @app.callback( Output("config-section", "style"), Input('selected-model', 'value') ) def toggle_config_section(selected_model): """ Toggles the visibility of the configuration section based on the selected model. Parameters: selected_model (str): The value of the selected model from the dropdown. Returns: dict: A dictionary containing the CSS style for the configuration section. """ if selected_model: return {"display": "block"} else: return {"display": "none"} @app.callback( Output('toggle-yaml', 'style'), Input('selected-model', 'value') ) def dipslay_yaml_container(selected_model): """ Toggles the display of the YAML update container based on the selected model. This function updates the CSS style of the YAML update container to either show or hide it based on whether a model is selected from the dropdown. Args: selected_model (str): The selected model from the dropdown. Returns: dict: A dictionary containing the CSS style to either display or hide the container. """ if selected_model: return {'display': 'block'} return {'display': 'none'} @app.callback( Output("offcanvas", "is_open"), [Input("open-offcanvas", "n_clicks")], [dash.dependencies.State("offcanvas", "is_open")], ) def toggle_offcanvas(n1, is_open): if n1: return not is_open return is_open @app.callback( Output("log-offcanvas", "is_open"), [Input("toggle-log", "n_clicks")], [State("log-offcanvas", "is_open")], ) def toggle_log(n, is_open): if n: return not is_open return is_open @app.callback( [Output('yaml-layout', 'style'), Output('yaml-layout', 'children')], [Input('modify-yaml-choice', 'value'), Input('selected-model', 'value')] ) def display_yaml_form(selection_update, selected_model): """ Toggles the display of the YAML form and updates its content based on user input. This function updates the CSS style and content of the YAML form based on whether the user chooses to update the YAML file and a model is selected from the dropdown. Args: selection_update (str): The user's choice to update the YAML file ('yes' or 'no'). selected_model (str): The selected model from the dropdown. Returns: tuple: A tuple containing the CSS style to either display or hide the form, and the form content generated from the YAML data. """ if not selected_model: return {'display': 'none'}, "Please select a model to display its configuration." try: yaml_conf = read_configs(selected_model) form_conf = create_yaml(yaml_conf) return {'display': 'block'}, form_conf except ValueError as e: return {'display': 'none'}, f"Error: {str(e)}" except Exception as e: return {'display': 'none'}, f"Unexpected Error: {str(e)}" @app.callback( Output("log-reader", "style"), Input('apply-button', 'n_clicks') ) def toggle_output_section(n_clicks): """ Toggles the visibility of the output logs section based on the number of clicks. Parameters: selected_model (str): The value of the selected model from the dropdown. Returns: dict: A dictionary containing the CSS style for the configuration section. """ if n_clicks is None or n_clicks == 0: return {'display': 'none'} return {'display': 'block'} @app.callback( Output('credentials-section', 'style'), [Input('modify-yaml-choice', 'value'), Input('selected-model', 'value'), Input('apply-button', 'n_clicks')] ) def display_credentials(selection_update, selected_model, n_clicks): """ Toggles the display of the credentials input fields based on user input. This function updates the CSS style of the credentials input fields to either show or hide them based on the user's choice to update the YAML file and the selection of a model from the dropdown. Args: selection_update (str): The user's choice to update the YAML file ('yes' or 'no'). selected_model (str): The selected model from the dropdown. Returns: dict: A dictionary containing the CSS style to either display or hide the credentials input fields. """ if n_clicks is None or n_clicks == 0: return {'display': 'none'} return {'display': 'block'} @app.callback( Output('process-button', 'style'), [Input('apply-button', 'n_clicks')] ) def display_launch_training(n_clicks): """ Displays the process button based on the number of clicks on the apply button. Parameters: n_clicks (int): The number of times the apply button has been clicked. Returns: dict: A dictionary containing the CSS style for the process button. """ if n_clicks and n_clicks > 0: return {'display': 'inline-block'} return {'display': 'none'} @app.callback( Output("results-section", "style"), Output("toggle-log", "className"), Input("process-button", "n_clicks") ) def display_results_section(n_clicks): """ Affiche la section des résultats et déclenche le clignotement du logo. """ if n_clicks and n_clicks > 0: return {"display": "block"}, "blinking" else: return {"display": "none"}, "" @app.callback( [Output('log-reader', 'children'), Output('acc-visualization', 'figure'), Output('acc-visualization', 'style'), Output('loss-visualization', 'figure'), Output('loss-visualization', 'style'), Output('confusion-matrix-img', 'src')], [Input('interval-widget', 'n_intervals'), Input('process-button', 'n_clicks')], [State('selected-model', 'value'), State('devcloud-username-input', 'value'), State('devcloud-password-input', 'value')] ) def refresh_metrics(n_intervals, nb_clicks, selected_model, devcloud_username, devcloud_password): """ Updates the log display and training metrics based on user actions and intervals. This function handles the following: - Executes the training script when the run button is clicked and updates the logs. - Periodically checks for new training metrics and updates the accuracy and loss graphs. - Manages the display of the log and metrics components based on the training status. Args: n_intervals (int): The number of intervals that have passed for the interval component. nb_clicks (int): The number of times the run button has been clicked. selected_model (str): The selected model from the dropdown. devcloud_username (str): The username for authentication. devcloud_password (str): The password for authentication. Returns: tuple: A tuple containing: - str: The updated log messages. - dict: The figure data for the accuracy graph. - dict: The CSS style to display or hide the accuracy graph. - dict: The figure data for the loss graph. - dict: The CSS style to display or hide the loss graph. - str: The base64 encoded image source for the confusion matrix. Raises: PreventUpdate: If the callback context is not triggered by a relevant input. """ global logs, new_training callback_context = dash.callback_context if not callback_context.triggered: raise PreventUpdate button = callback_context.triggered[0]['prop_id'].split('.')[0] if button == 'process-button' and nb_clicks: if devcloud_username and devcloud_password: st_script = f"stm32ai-modelzoo-services/{selected_model}/src/stm32ai_main.py" execute_async(st_script, devcloud_username, devcloud_password) new_training = True logs.append("Starting application ...") return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None else: logs.append("Please enter both ST Developer Cloud username and password:") return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None elif button == 'interval-widget': if not new_training: return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None outputs_folder = "experiments_outputs" if not os.path.exists(outputs_folder): os.makedirs(outputs_folder) return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None dated_directories = [d for d in os.listdir(outputs_folder) if os.path.isdir(os.path.join(outputs_folder, d)) and d.startswith('20')] if dated_directories: recent_directory = max(dated_directories, key=lambda d: datetime.strptime(d, '%Y_%m_%d_%H_%M_%S')) train_metrics_file = os.path.join(outputs_folder, recent_directory, 'logs', 'metrics', 'train_metrics.csv') print(f"Metrics file : {train_metrics_file}") if os.path.exists(train_metrics_file) and new_training: metrics_dataframe = pd.read_csv(train_metrics_file) if not metrics_dataframe.empty: figures = [] metrics_pairs = [ ('accuracy', 'val_accuracy'), ('loss', 'val_loss'), ('oks', 'val_oks'), ('val_map',) ] for pair in metrics_pairs: if len(pair) == 2: train_metric, val_metric = pair if train_metric in metrics_dataframe.columns and val_metric in metrics_dataframe.columns: fig = { 'data': [ { 'x': metrics_dataframe['epoch'], 'y': metrics_dataframe[train_metric], 'type': 'line', 'name': train_metric.capitalize(), 'line': {'color': '#FFD200', 'width': 2, 'dash': 'solid'}, 'hoverinfo': 'x+y+name', 'hoverlabel': {'bgcolor': '#EEEFF1', 'font': {'color': '#525A63'}} }, { 'x': metrics_dataframe['epoch'], 'y': metrics_dataframe[val_metric], 'type': 'line', 'name': val_metric.capitalize(), 'line': {'color': '#3CB4E6', 'width': 2, 'dash': 'solid'}, 'hoverinfo': 'x+y+name', 'hoverlabel': {'bgcolor': '#EEEFF1', 'font': {'color': '#525A63'}} } ], 'layout': { 'title': { 'text': f'{train_metric.capitalize()} vs {val_metric.capitalize()}', 'x': 0.5, 'xanchor': 'center' }, 'xaxis': { 'title': 'Epochs', 'showgrid': True, 'gridcolor': '#EEEFF1', 'tickangle': 45 }, 'yaxis': { 'title': train_metric.capitalize(), 'showgrid': True, 'gridcolor': '#EEEFF1' }, 'showlegend': True, 'legend': { 'x': 1, 'y': 1, 'traceorder': 'normal', 'font': {'size': 10}, 'bgcolor': '#EEEFF1', 'bordercolor': '#A6ADB5', 'borderwidth': 1 }, 'hovermode': 'closest', 'plot_bgcolor': '#ffffff' } } figures.append(fig) elif len(pair) == 1: val_metric = pair[0] if val_metric in metrics_dataframe.columns: fig = { 'data': [ { 'x': metrics_dataframe['epoch'], 'y': metrics_dataframe[val_metric], 'type': 'line', 'name': val_metric.capitalize(), 'line': {'color': '#3CB4E6', 'width': 2, 'dash': 'solid'}, 'hoverinfo': 'x+y+name', 'hoverlabel': {'bgcolor': '#EEEFF1', 'font': {'color': '#525A63'}} } ], 'layout': { 'title': { 'text': f'{val_metric.capitalize()} over Epochs', 'x': 0.5, 'xanchor': 'center' }, 'xaxis': { 'title': 'Epochs', 'showgrid': True, 'gridcolor': '#EEEFF1', 'tickangle': 45 }, 'yaxis': { 'title': val_metric.capitalize(), 'showgrid': True, 'gridcolor': '#EEEFF1' }, 'showlegend': True, 'legend': { 'x': 1, 'y': 1, 'traceorder': 'normal', 'font': {'size': 10}, 'bgcolor': '#EEEFF1', 'bordercolor': '#A6ADB5', 'borderwidth': 1 }, 'hovermode': 'closest', 'plot_bgcolor': '#ffffff' } } figures.append(fig) confusion_matrix_src = latest_confusion_matrix(outputs_folder, recent_directory) if figures: return "\n".join(logs), figures[0], {'display': 'block'}, figures[1] if len(figures) > 1 else {}, {'display': 'block'}, confusion_matrix_src else: return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, confusion_matrix_src else: return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None else: return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None else: return "\n".join(logs), {}, {'display': 'none'}, {}, {'display': 'none'}, None raise PreventUpdate @app.callback( Output('metric-graphs-container', 'children'), Input('log-reader', 'children') ) def update_metrics_dashboard(logs): metrics = extract_metrics(logs) graphs = [] def get_metric_card(title, figure): return dbc.Col( dbc.Card([ dbc.CardHeader(title, style={'background-color': '#8191a5', 'color': 'white', 'font-size': '18px'}), dbc.CardBody(dcc.Graph(figure=figure, config={'displayModeBar': False})) ]), width=4 ) average_iou = metrics.get("float", {}).get("average_iou") if average_iou is not None: graphs.append(get_metric_card("Average IoU - Float Model", create_iou_gauge(average_iou))) if "float" in metrics: acc = metrics["float"].get("accuracy") if acc is not None: graphs.append(get_metric_card("Accuracy - Float Model", create_accuracy_gauge(acc))) if "quantized" in metrics: acc = metrics["quantized"].get("accuracy") if acc is not None: graphs.append(get_metric_card("Accuracy - Quantized Model", create_accuracy_gauge(acc))) if "oks" in metrics: mean_oks = metrics["oks"].get("mean_oks") if mean_oks is not None: graphs.append(get_metric_card("Mean OKS", create_accuracy_gauge(mean_oks))) precision = metrics["oks"].get("precision") if precision is not None: graphs.append(get_metric_card("Mean Precision", create_accuracy_gauge(precision))) recall = metrics["oks"].get("recall") if recall is not None: graphs.append(get_metric_card("Mean Recall", create_accuracy_gauge(recall))) mean_ap = metrics["oks"].get("mean_ap") if mean_ap is not None: graphs.append(get_metric_card("Mean AP (mAP)", create_accuracy_gauge(mean_ap))) return dbc.Row(graphs, justify="center") @app.callback( Output('memory-bar', 'figure'), Input('log-reader', 'children') ) def update_memory_bar(logs): metrics = _parse_inference_memory(logs) ram = metrics.get('ram', 0) flash = metrics.get('flash', 0) fig = go.Figure() fig.add_trace(go.Bar( y=["Total RAM ", "Total Flash"], x=[ram, flash], orientation='h', marker_color=["#E6007E", "#3cb4e6"] )) fig.update_layout(title="Memory Usage (KiB)", xaxis_title="Size (KiB)") return fig @app.callback( Output('inference-time', 'figure'), Input('log-reader', 'children') ) def update_inference_time(logs): metrics = _parse_inference_memory(logs) inference_time = metrics.get('inference_time', 0) fig = go.Figure(go.Indicator( mode="number", value=inference_time, title={'text': "Inference Time (ms)"}, )) return fig @app.callback( Output('submission-outcome', 'children'), [Input('apply-button', 'n_clicks'), Input('process-button', 'n_clicks')], [State({'type': 'yaml-setting', 'index': ALL}, 'id'), State({'type': 'yaml-setting', 'index': ALL}, 'value'), State('selected-model', 'value'), State('devcloud-username-input', 'value'), State('devcloud-password-input', 'value')] ) def process_button_actions(submit_clicks, exec_nb_clicks, form_input_ids, form_input_values, selected_model, devcloud_username, devcloud_password): """ Handles the actions triggered by the submit and run buttons. This function processes the form data when the submit button is clicked, updates the corresponding YAML file, and executes the training script when the run button is clicked. Args: submit_clicks (int): The number of times the submit button has been clicked. exec_nb_clicks (int): The number of times the execution/run button has been clicked. form_input_ids (list): A list of dictionaries containing the IDs of the form inputs. form_input_values (list): A list of values from the form inputs. selected_model (str): The selected model from the dropdown. devcloud_username (str): The username for DevCloud authentication. devcloud_password (str): The password for DevCloud authentication. Returns: str: A message indicating the result of the action, such as successful YAML update or script execution status. Raises: PreventUpdate: If the callback context is not triggered by a relevant input or if no action is taken. """ new_fields = [] callback_context = dash.callback_context if not callback_context.triggered: raise PreventUpdate triggered_button = callback_context.triggered[0]['prop_id'].split('.')[0] if triggered_button == 'apply-button': if submit_clicks: try: form_fields_data = {} for i in range(len(form_input_ids)): input_id = form_input_ids[i]['index'] input_value = form_input_values[i] form_fields_data[input_id] = input_value yaml_file_path = local_yamls.get(selected_model) if yaml_file_path : yaml_parser = ruamel.yaml.YAML() with open(yaml_file_path , 'r') as file: current_yaml_data = yaml_parser.load(file) updated_yaml_data = process_form_configs(form_fields_data) for key, value in updated_yaml_data.items(): keys = key.split('.') nested_dict = current_yaml_data for k in keys[:-1]: nested_dict = nested_dict.setdefault(k, {}) if nested_dict[keys[-1]] != value: nested_dict[keys[-1]] = value new_fields.append(key) with open(yaml_file_path , 'w') as file: yaml_parser.dump(current_yaml_data, file) return f"User config yaml file has been updated successfully ! Updated fields are: {', '.join(new_fields)}" else: return f"ERROR: No user config yaml found for '{selected_model}'." except Exception as e: return f"ERROR: UPDATING USER CONFIG YAML file: {e}" else: raise PreventUpdate elif triggered_button == 'process-button': if exec_nb_clicks: st_script = f"stm32ai-modelzoo-services/{selected_model}/src/stm32ai_main.py" execute_async(st_script, devcloud_username, devcloud_password) return "Application is running ..." else: raise PreventUpdate @app.callback( Output('download-action', 'style'), [Input('interval-widget', 'n_intervals')], [State('selected-model', 'value')] ) def toggle_download_button(n_intervals, selected_model): """ Toggles the display of the download button based on the existence of output directories. This function checks if the output directories for the selected model exist and toggles the display of the download button accordingly. Args: n_intervals (int): The number of intervals that have passed for the interval component. model_choice (str): The selected model from the dropdown. Returns: dict: A dictionary containing the CSS style to either display or hide the download button. """ out_directory = os.path.join(os.getcwd(), "experiments_outputs") if not os.path.exists(out_directory ): return {'display': 'none'} output_subdirectories = [d for d in os.listdir(out_directory ) if os.path.isdir(os.path.join(out_directory , d)) and d.startswith('20')] if output_subdirectories: return {'display': 'block'} return {'display': 'none'} @app.callback( Output('download-resource', 'data'), [Input('download-action', 'n_clicks')], [State('selected-model', 'value')] ) def generate_download_link(n_clicks, selected_model): """ Generates a download link based on the selected model and operation mode. This function reads the YAML configuration for the selected model, determines the operation mode, and generates a download link for the appropriate file (ZIP or ELF/BIN) based on the operation mode. Args: click_count (int): The number of times the download button has been clicked. selected_model (str): The selected model from the dropdown. Returns: dcc.send_file: A Dash component to send the file for download. Raises: PreventUpdate: If no relevant action is taken or the required files do not exist. """ if n_clicks is None: raise PreventUpdate output_directory = os.path.join(os.getcwd(), "./experiments_outputs") if not os.path.exists(output_directory ): raise PreventUpdate timestamped_directories = [d for d in os.listdir(output_directory ) if os.path.isdir(os.path.join(output_directory , d)) and d.startswith('20')] timestamped_directories = [ d for d in os.listdir(output_directory) if os.path.isdir(os.path.join(output_directory, d)) and d.startswith("20") ] if timestamped_directories: recent_directory = max( timestamped_directories, key=lambda d: datetime.strptime(d, "%Y_%m_%d_%H_%M_%S") ) recent_directory_path = os.path.join(output_directory, recent_directory) zip_file_path = os.path.join(recent_directory_path, f"{recent_directory}.zip") if not os.path.exists(zip_file_path): create_archive(zip_file_path, recent_directory_path) if os.path.exists(zip_file_path): return dcc.send_file(zip_file_path) raise PreventUpdate @server.route('/download/') def download_file(subpath): """ Route to download a file from the server. Parameters: - subpath (str): The subpath of the file to be downloaded, relative to the './experiments_outputs' directory. Returns: - Response: A Flask response object to send the file as an attachment if it exists. - tuple: A tuple containing an error message and a 404 status code if the file is not found. """ file_path = os.path.join(os.getcwd(), './experiments_outputs', subpath) if os.path.exists(file_path): return send_file(file_path, as_attachment=True) else: return "File not found", 404 if __name__ == '__main__': app.run_server(host='0.0.0.0',port=7860, dev_tools_ui=True, dev_tools_hot_reload=True, threaded=True)