rogerkoranteng's picture
Upload folder using huggingface_hub
3506b46 verified
import os
import json
import torch
import sqlite3
import gradio as gr
import faiss
import pandas as pd
from datetime import datetime
from PIL import Image
from flask import Flask, jsonify, request
from werkzeug.utils import secure_filename
from threading import Thread
from transformers import AutoFeatureExtractor, AutoModelForImageClassification, AutoTokenizer, AutoModel, \
AutoModelForCausalLM
from matplotlib import pyplot as plt
from sklearn.linear_model import LinearRegression
import numpy as np
import seaborn as sns
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
from torch import nn, optim
from torch.optim import lr_scheduler
from sklearn.preprocessing import LabelEncoder
# Initialize Flask app and models
app = Flask(__name__, static_folder="static")
# Constants
LOW_STOCK_THRESHOLD = 5 # Customize threshold as needed
DATABASE = 'uploaded_images.db'
# Available model options
MODEL_OPTIONS = {
"Google ViT (Base)": "google/vit-base-patch16-224",
"Google ViT (Large)": "google/vit-large-patch16-224",
"Microsoft ResNet50": "microsoft/resnet-50",
"Facebook ConvNeXt Tiny": "facebook/convnext-tiny-224",
"Microsoft Swin": "microsoft/swin-tiny-patch4-window7-224",
}
# Set default model
selected_model_name = MODEL_OPTIONS["Google ViT (Base)"]
feature_extractor = AutoFeatureExtractor.from_pretrained(selected_model_name)
model = AutoModelForImageClassification.from_pretrained(selected_model_name)
class_names = model.config.id2label
# Initialize inventory data
inventory_data = {}
# Initialize database
def init_db():
conn = sqlite3.connect(DATABASE)
cursor = conn.cursor()
cursor.execute("""
CREATE TABLE IF NOT EXISTS images (
id INTEGER PRIMARY KEY AUTOINCREMENT,
filename TEXT,
upload_time TEXT
)
""")
conn.commit()
conn.close()
init_db()
def plot_bar_chart(inventory_data):
# Get the list of items and their counts
items = list(inventory_data.keys())
counts = [inventory_data[item]["count"] for item in items]
# Extract only the first word from each item name
first_words = [item.split()[0] for item in items]
# Create the bar chart
fig, ax = plt.subplots(figsize=(8, 6))
ax.bar(first_words, counts, color="skyblue")
# Set titles and labels
ax.set_title("Inventory Counts")
ax.set_xlabel("Items")
ax.set_ylabel("Count")
# Save the chart as a PNG file
chart_path = "bar_chart.png"
plt.tight_layout()
plt.savefig(chart_path)
plt.close()
return chart_path
def plot_line_chart(inventory_data):
# Extract items, counts, and dates from inventory data
items = list(inventory_data.keys())
counts = [inventory_data[item]["count"] for item in items]
dates = [inventory_data[item]["last_detected"] for item in items]
# Convert string dates to pandas datetime objects
dates = pd.to_datetime(dates)
# Create a DataFrame with dates and counts
counts_df = pd.DataFrame({"Date": dates, "Count": counts})
counts_df.sort_values("Date", inplace=True)
# Format the 'Date' column into a human-readable format
counts_df["Date"] = counts_df["Date"].dt.strftime('%Y-%m-%d %H:%M:%S')
# Create the plot
fig, ax = plt.subplots(figsize=(8, 6))
ax.plot(counts_df["Date"], counts_df["Count"], marker="o", color="orange")
# Set the title and labels for the plot
ax.set_title("Stock Changes Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Stock Level")
# Save the plot as a PNG file
chart_path = "line_chart.png"
plt.xticks(rotation=45) # Rotate x-ticks for better readability
plt.tight_layout()
plt.savefig(chart_path)
plt.close()
return chart_path
def plot_pie_chart(inventory_data):
# Get the list of items and their counts
items = list(inventory_data.keys())
counts = [inventory_data[item]["count"] for item in items]
# Extract only the first word from each item name
first_words = [item.split()[0] for item in items]
# Create the pie chart
fig, ax = plt.subplots(figsize=(8, 6))
ax.pie(counts, labels=first_words, autopct='%1.1f%%', startangle=140, colors=plt.cm.Paired.colors)
# Set the title
ax.set_title("Product Category Breakdown")
# Save the chart as a PNG file
chart_path = "pie_chart.png"
plt.tight_layout()
plt.savefig(chart_path)
plt.close()
return chart_path
def plot_heatmap(inventory_data):
# Get the list of items
items = list(inventory_data.keys())
# Extract only the first word from each item name
first_words = [item.split()[0] for item in items]
# Create a matrix where the count is placed at the correct location
change_matrix = [[inventory_data[item]["count"] if item == other else 0 for other in items] for item in items]
# Create the heatmap
fig, ax = plt.subplots(figsize=(8, 6))
cax = ax.imshow(change_matrix, cmap="YlOrRd", interpolation='nearest')
# Set the labels on the x and y axes to the first words of the items
ax.set_xticks(range(len(items)))
ax.set_yticks(range(len(items)))
ax.set_xticklabels(first_words)
ax.set_yticklabels(first_words)
# Add a colorbar
fig.colorbar(cax)
# Set the title of the heatmap
ax.set_title("Stock Change Heatmap")
# Save the chart as a PNG file
chart_path = "heatmap.png"
plt.tight_layout()
plt.savefig(chart_path)
plt.close()
return chart_path
# Function to return the paths of the images
def generate_bar_chart():
return plot_bar_chart(inventory_data)
def generate_line_chart():
return plot_line_chart(inventory_data)
def generate_pie_chart():
return plot_pie_chart(inventory_data)
def generate_heatmap():
return plot_heatmap(inventory_data)
# Utility functions
def log_inventory_data(item_class):
try:
timestamp = datetime.now().isoformat()
# Initialize the inventory data for the item class if it's not present
if item_class not in inventory_data:
inventory_data[item_class] = {
"category": item_class,
"count": 0, # Initial count
"last_detected": None, # Timestamp of last detection
"history": [] # Track the historical counts over time
}
# Log the count and timestamp to the history
inventory_data[item_class]["history"].append({
"timestamp": timestamp, # Store the timestamp as a string
"count": inventory_data[item_class]["count"]
})
# Update the count of the item class
inventory_data[item_class]["count"] += 1
# Update the last detected timestamp
inventory_data[item_class]["last_detected"] = timestamp
# Optionally: Save the data to a file for persistence
with open("inventory_log.json", "w") as f:
json.dump(inventory_data, f, indent=4)
print(f"Inventory data logged for item class: {item_class}")
except Exception as e:
print(f"Error logging inventory data: {e}")
def check_stock_levels():
try:
for item, details in inventory_data.items():
if details["count"] < LOW_STOCK_THRESHOLD:
print(f"Low stock detected for {item}: {details['count']} items.")
except Exception as e:
print(f"Error checking stock levels: {e}")
def save_image_to_db(file):
filename = secure_filename(file.filename)
upload_time = datetime.now().isoformat()
file_path = os.path.join("uploads", filename)
file.save(file_path)
conn = sqlite3.connect(DATABASE)
cursor = conn.cursor()
cursor.execute("INSERT INTO images (filename, upload_time) VALUES (?, ?)", (filename, upload_time))
conn.commit()
conn.close()
def batch_predict(images):
results = []
try:
for image_file in images:
with Image.open(image_file.name) as image:
if image.mode != "RGB":
image = image.convert("RGB")
inputs = feature_extractor(images=image, return_tensors="pt")
outputs = model(**inputs)
predicted_class_id = torch.argmax(outputs.logits, dim=1).item()
item_class = class_names[predicted_class_id]
log_inventory_data(item_class) # Pass only the item_class argument
results.append({"Image": os.path.basename(image_file.name), "Classification": item_class})
except Exception as e:
return [{"Image": "Error", "Classification": str(e)}]
return pd.DataFrame(results)
def forecast_inventory(item_class, days=7):
"""Use linear regression for basic inventory forecasting."""
try:
# Get the historical data for the item class
if item_class not in inventory_data or not inventory_data[item_class]["history"]:
return {"error": f"No historical data for {item_class}."}
history = inventory_data[item_class]["history"]
timestamps = [datetime.fromisoformat(entry["timestamp"]) for entry in history]
counts = [entry["count"] for entry in history]
# Convert timestamps to days since the first entry
days_since_first_entry = [(timestamp - timestamps[0]).days for timestamp in timestamps]
# Apply linear regression to forecast the next `days` values
model = LinearRegression()
model.fit(np.array(days_since_first_entry).reshape(-1, 1), counts)
# Predict future inventory counts
future_days = np.array([days_since_first_entry[-1] + i for i in range(1, days + 1)]).reshape(-1, 1)
predictions = model.predict(future_days)
forecast = [{"day": i + 1, "predicted_count": int(predictions[i])} for i in range(days)]
return forecast
print(f"Inventory data logged for item class: {forecast}")
except Exception as e:
return {"error": f"Error in forecasting: {str(e)}"}
def change_model(selected_model_key):
try:
global feature_extractor, model, class_names
if selected_model_key not in MODEL_OPTIONS:
return "Error: Invalid model selection"
# Retrieve the model path (either pre-trained or fine-tuned)
selected_model_path = MODEL_OPTIONS[selected_model_key]
# Load the model from the specified path
feature_extractor = AutoFeatureExtractor.from_pretrained(selected_model_path)
model = AutoModelForImageClassification.from_pretrained(selected_model_path)
class_names = model.config.id2label
return f"Model changed to {selected_model_key}"
except Exception as e:
return f"Error changing model: {str(e)}"
# Define a custom dataset class
class CustomDataset(Dataset):
def __init__(self, images, labels, transform=None):
self.images = images
self.labels = labels
self.transform = transform
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
image = Image.open(self.images[idx]).convert("RGB")
label = self.labels[idx]
if self.transform:
image = self.transform(image)
return image, label
# Define transformations for image preprocessing
transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
# Function to preprocess and load data
def preprocess_data(train_images, train_labels):
dataset = CustomDataset(train_images, train_labels, transform=transform)
dataloader = DataLoader(dataset, batch_size=16, shuffle=True)
return dataloader
# Fine-tune the model with a custom dataset
def fine_tune_model(train_images, train_labels, model_name="custom_model"):
try:
# Convert image paths to a suitable format
train_images = [image.name for image in train_images] # list of image paths
# Check if labels need encoding
if isinstance(train_labels, str):
train_labels = train_labels.split(",") # Convert the comma-separated string to a list
# Encode labels if they are not integers
label_encoder = LabelEncoder()
train_labels = label_encoder.fit_transform(train_labels)
# Load the model and prepare for fine-tuning
global model
model = AutoModelForImageClassification.from_pretrained(selected_model_name)
# Modify the classifier layer for the new dataset
num_labels = len(set(train_labels)) # Number of unique labels
model.classifier = nn.Linear(model.config.hidden_size, num_labels)
# Prepare data loader
dataloader = preprocess_data(train_images, train_labels)
# Define loss function, optimizer, and learning rate scheduler
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.0001)
scheduler = lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1)
# Training loop
num_epochs = 50 # Set the number of epochs
model.train() # Set model to training mode
for epoch in range(num_epochs):
running_loss = 0.0
for inputs, labels in dataloader:
optimizer.zero_grad()
# Forward pass
outputs = model(inputs)
logits = outputs.logits if hasattr(outputs, "logits") else outputs # Handle outputs
# Calculate loss
loss = criterion(logits, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
scheduler.step()
print(f"Epoch {epoch + 1}/{num_epochs}, Loss: {running_loss / len(dataloader)}")
# return f"Epoch {epoch + 1}/{num_epochs}, Loss: {running_loss / len(dataloader)}"
# Save the fine-tuned model
fine_tuned_model_path = f"models/{model_name}" # Save path
model.save_pretrained(fine_tuned_model_path)
print(f"Fine-tuned model saved to {fine_tuned_model_path}")
# Dynamically add the fine-tuned model to the model selection
MODEL_OPTIONS[model_name] = fine_tuned_model_path
return f"Fine-tuned model {model_name} has been added successfully!"
except Exception as e:
return f"Error fine-tuning the model: {str(e)}"
proxy_prefix = os.environ.get("PROXY_PREFIX")
# Gradio Interface
with gr.Blocks() as interface:
gr.Markdown("## VisionTrack - Smart Inventory Management and Analysis System")
with gr.Tab("Model Selection/Fine-Tuning"):
with gr.Tab("Select Model"):
gr.Markdown("Choose a model for image classification.")
model_dropdown = gr.Dropdown(choices=list(MODEL_OPTIONS.keys()), label="Select Model",
value="Google ViT (Base)")
model_dropdown.change(change_model, inputs=model_dropdown, outputs=gr.Textbox(label="status"))
with gr.Tab("Model Fine-Tuning"):
gr.Markdown("Upload your custom dataset for model fine-tuning.")
train_image_input = gr.File(label="Upload Training Images", file_count="multiple", type="filepath")
train_label_input = gr.Textbox(label="Enter Corresponding Labels (comma-separated)")
fine_tune_button = gr.Button("Fine-Tune Model")
fine_tune_output = gr.Text(label="Training Status")
fine_tune_button.click(fine_tune_model, inputs=[train_image_input, train_label_input],
outputs=fine_tune_output)
with gr.Tab("Image Classification"):
gr.Markdown("Upload images for classification.")
image_input = gr.File(label="Upload Images", file_count="multiple", type="filepath")
output = gr.DataFrame(label="Classification Results", interactive=True)
image_input.upload(batch_predict, inputs=image_input, outputs=output)
with gr.Tab("Inventory"):
gr.Markdown("Check current inventory.")
inventory_display = gr.DataFrame(value=pd.DataFrame(columns=["Item", "Count", "Last Detected"]))
gr.Button("Fetch Inventory").click(lambda: pd.DataFrame(inventory_data).T, outputs=inventory_display)
with gr.Tab("Forecasting"):
gr.Markdown("Forecast inventory levels for the next 7 days.")
item_class_input = gr.Textbox(label="Enter Item Class")
forecast_button = gr.Button("Forecast")
forecast_output = gr.Text(label="Predicted Inventory Forecast")
forecast_button.click(forecast_inventory, inputs=item_class_input, outputs=forecast_output)
with gr.Tab("Inventory Dashboard"):
with gr.Tab("Bar Chart"):
bar_chart_image = gr.Image(type="filepath", label="Bar Chart")
gr.Button("Generate Bar Chart").click(generate_bar_chart, outputs=bar_chart_image)
with gr.Tab("Line Chart"):
line_chart_image = gr.Image(type="filepath", label="Line Chart")
gr.Button("Generate Line Chart").click(generate_line_chart, outputs=line_chart_image)
with gr.Tab("Pie Chart"):
pie_chart_image = gr.Image(type="filepath", label="Pie Chart")
gr.Button("Generate Pie Chart").click(generate_pie_chart, outputs=pie_chart_image)
with gr.Tab("Heatmap"):
heatmap_image = gr.Image(type="filepath", label="Heatmap")
gr.Button("Generate Heatmap").click(generate_heatmap, outputs=heatmap_image)
if __name__ == '__main__':
app_thread = Thread(target=lambda: app.run(debug=True, use_reloader=False))
app_thread.start()
interface.launch(server_name="0.0.0.0", server_port=8080, root_path=proxy_prefix, share=True)