text2svg-demo-app / eval_analysis.py
Jinglong Xiong
add analysis script
c0f4df5
raw
history blame contribute delete
10.6 kB
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import json
from pathlib import Path
# Set style
plt.style.use('ggplot')
sns.set_palette("Set2")
plt.rcParams['figure.figsize'] = (12, 8)
# Load the data
results_csv = "results/summary_20250421_230054.csv"
results_json = "results/results_20250421_230054.json"
df = pd.read_csv(results_csv)
# Extract category from description if not already available
def extract_category(row):
"""
Determines the category of an image based on its description or existing category.
Args:
row: A pandas DataFrame row containing 'category' and 'description' fields
Returns:
str: The determined category ('fashion', 'landscape', 'abstract', or 'unknown')
"""
if pd.notna(row['category']) and row['category'] != 'unknown':
return row['category']
# Try to extract from description
desc = row['description'].lower()
if any(keyword in desc for keyword in ['coat', 'pants', 'shirt', 'dress', 'scarf', 'shoes']):
return 'fashion'
elif any(keyword in desc for keyword in ['forest', 'beach', 'mountain', 'ocean', 'lake', 'sky']):
return 'landscape'
elif any(keyword in desc for keyword in ['rectangle', 'circle', 'triangle', 'shape', 'spiral']):
return 'abstract'
else:
return 'unknown'
# Clean the data
df['category'] = df.apply(extract_category, axis=1)
df['generation_time'] = pd.to_numeric(df['generation_time'], errors='coerce')
# 1. Model Performance Comparison
def plot_model_comparison():
"""
Creates boxplots comparing model performance across three metrics:
VQA score, aesthetic score, and fidelity score.
Saves the resulting plot to 'results/model_comparison.png'.
"""
fig, axes = plt.subplots(1, 3, figsize=(18, 6))
metrics = ['vqa_score', 'aesthetic_score', 'fidelity_score']
titles = ['VQA Score', 'Aesthetic Score', 'Fidelity Score']
for i, (metric, title) in enumerate(zip(metrics, titles)):
sns.boxplot(x='model', y=metric, data=df, ax=axes[i])
axes[i].set_title(f'{title} by Model')
axes[i].set_ylim([0, 1])
plt.tight_layout()
plt.savefig('results/model_comparison.png')
plt.close()
# 2. Category Performance Analysis
def plot_category_performance():
"""
Creates boxplots showing performance by category and model for three metrics:
VQA score, aesthetic score, and fidelity score.
Saves the resulting plot to 'results/category_performance.png'.
"""
fig, axes = plt.subplots(1, 3, figsize=(18, 6))
metrics = ['vqa_score', 'aesthetic_score', 'fidelity_score']
titles = ['VQA Score', 'Aesthetic Score', 'Fidelity Score']
for i, (metric, title) in enumerate(zip(metrics, titles)):
sns.boxplot(x='category', y=metric, hue='model', data=df, ax=axes[i])
axes[i].set_title(f'{title} by Category and Model')
axes[i].set_ylim([0, 1])
if i > 0:
axes[i].get_legend().remove()
axes[0].legend(title='Model')
plt.tight_layout()
plt.savefig('results/category_performance.png')
plt.close()
# 3. Generation Time Analysis
def plot_generation_time():
"""
Creates visualizations of generation time analysis:
1. A boxplot showing generation time by model
2. Scatter plots showing the relationship between generation time and quality metrics
Saves the resulting plots to 'results/generation_time.png' and 'results/quality_vs_time.png'.
"""
plt.figure(figsize=(10, 6))
sns.boxplot(x='model', y='generation_time', data=df)
plt.title('Generation Time by Model')
plt.ylabel('Time (seconds)')
plt.tight_layout()
plt.savefig('results/generation_time.png')
plt.close()
# Generation time vs quality scatter plot
fig, axes = plt.subplots(1, 3, figsize=(18, 6))
metrics = ['vqa_score', 'aesthetic_score', 'fidelity_score']
titles = ['VQA Score', 'Aesthetic Score', 'Fidelity Score']
for i, (metric, title) in enumerate(zip(metrics, titles)):
for model, color in zip(df['model'].unique(), ['#1f77b4', '#ff7f0e']):
model_data = df[df['model'] == model]
axes[i].scatter(model_data['generation_time'], model_data[metric],
alpha=0.6, label=model, c=color)
axes[i].set_title(f'{title} vs. Generation Time')
axes[i].set_xlabel('Generation Time (seconds)')
axes[i].set_ylabel(title)
axes[i].legend()
plt.tight_layout()
plt.savefig('results/quality_vs_time.png')
plt.close()
# 4. Description complexity vs performance
def plot_complexity_performance():
"""
Analyzes the relationship between description complexity (word count) and
performance metrics, creating scatter plots with trend lines.
Saves the resulting plot to 'results/complexity_performance.png'.
"""
df['description_length'] = df['description'].str.len()
df['word_count'] = df['description'].str.split().str.len()
fig, axes = plt.subplots(1, 3, figsize=(18, 6))
metrics = ['vqa_score', 'aesthetic_score', 'fidelity_score']
titles = ['VQA Score', 'Aesthetic Score', 'Fidelity Score']
for i, (metric, title) in enumerate(zip(metrics, titles)):
for model, color in zip(df['model'].unique(), ['#1f77b4', '#ff7f0e']):
model_data = df[df['model'] == model]
axes[i].scatter(model_data['word_count'], model_data[metric],
alpha=0.6, label=model, c=color)
# Add trendline
z = np.polyfit(model_data['word_count'], model_data[metric], 1)
p = np.poly1d(z)
axes[i].plot(sorted(model_data['word_count']), p(sorted(model_data['word_count'])),
c=color, linestyle='--')
axes[i].set_title(f'{title} vs. Description Complexity')
axes[i].set_xlabel('Word Count')
axes[i].set_ylabel(title)
axes[i].legend()
plt.tight_layout()
plt.savefig('results/complexity_performance.png')
plt.close()
# 5. Success and failure examples
def analyze_best_worst_examples():
"""
Identifies and prints the top 10 most successful and least successful generations
based on fidelity score.
Creates directories for sample SVG and PNG files if they don't exist.
Returns:
tuple: (success_df, failure_df) DataFrames containing the best and worst examples
"""
# Create directory for result samples
Path("results/sample_svg").mkdir(exist_ok=True)
Path("results/sample_png").mkdir(exist_ok=True)
# Load detailed results
with open(results_json, 'r') as f:
results_data = json.load(f)
# Create success/failure dataframes
success_df = df.nlargest(10, 'fidelity_score')
failure_df = df.nsmallest(10, 'fidelity_score')
# Print success examples
print("Top 10 Successful Generations:")
print(success_df[['model', 'description', 'vqa_score', 'aesthetic_score', 'fidelity_score']].to_string(index=False))
# Print failure examples
print("\nTop 10 Failed Generations:")
print(failure_df[['model', 'description', 'vqa_score', 'aesthetic_score', 'fidelity_score']].to_string(index=False))
return success_df, failure_df
# 6. Summary statistics
def print_summary_stats():
"""
Calculates and prints summary statistics for model performance:
1. Overall stats by model (mean, std, min, max for each metric)
2. Performance by category and model
Also creates a radar chart visualizing fidelity scores by category and model,
saved to 'results/category_radar.png'.
"""
# Overall stats by model
model_stats = df.groupby('model').agg({
'vqa_score': ['mean', 'std', 'min', 'max'],
'aesthetic_score': ['mean', 'std', 'min', 'max'],
'fidelity_score': ['mean', 'std', 'min', 'max'],
'generation_time': ['mean', 'std', 'min', 'max']
})
print("Overall Model Performance:")
print(model_stats)
# Stats by category and model
category_stats = df.groupby(['model', 'category']).agg({
'vqa_score': 'mean',
'aesthetic_score': 'mean',
'fidelity_score': 'mean',
'generation_time': 'mean'
}).reset_index()
print("\nPerformance by Category and Model:")
print(category_stats.to_string())
# Create a radar chart for category performance
categories = category_stats['category'].unique()
models = category_stats['model'].unique()
plt.figure(figsize=(10, 8))
angles = np.linspace(0, 2*np.pi, len(categories), endpoint=False).tolist()
angles += angles[:1] # Close the loop
ax = plt.subplot(111, polar=True)
for model in models:
model_data = category_stats[category_stats['model'] == model]
values = []
for category in categories:
cat_data = model_data[model_data['category'] == category]
if not cat_data.empty:
values.append(cat_data['fidelity_score'].values[0])
else:
values.append(0)
values += values[:1] # Close the loop
ax.plot(angles, values, linewidth=2, label=model)
ax.fill(angles, values, alpha=0.25)
ax.set_xticks(angles[:-1])
ax.set_xticklabels(categories)
ax.set_title('Fidelity Score by Category and Model')
ax.legend(loc='upper right')
plt.tight_layout()
plt.savefig('results/category_radar.png')
plt.close()
# Main analysis function
def run_analysis():
"""
Main function that runs the complete analysis pipeline:
1. Creates necessary directories
2. Generates all visualization plots
3. Prints summary statistics
4. Analyzes best and worst examples
All results are saved to the 'results/' directory.
"""
print("Starting analysis of evaluation results...")
# Create plots directory if it doesn't exist
Path("results").mkdir(exist_ok=True)
# Generate all plots
plot_model_comparison()
plot_category_performance()
plot_generation_time()
plot_complexity_performance()
# Print summary statistics
print_summary_stats()
# Analyze best and worst examples
success_df, failure_df = analyze_best_worst_examples()
print("\nAnalysis complete. Visualizations saved to 'results/' directory.")
if __name__ == "__main__":
run_analysis()