File size: 16,466 Bytes
c6535db | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 | """Experiment logger for FSampler research data collection."""
import os
import json
import csv
import time
from datetime import datetime
import torch
import numpy as np
from PIL import Image
import folder_paths
class ExperimentLogger:
"""Handles saving FSampler experiment runs to disk with all metadata and outputs."""
def __init__(self, base_path=None):
"""
Initialize the experiment logger.
Args:
base_path: Base directory for experiments (defaults to ComfyUI/output/experiments/)
If empty string, uses default. If provided, uses that exact path.
"""
if base_path is None or base_path == "":
# Default to ComfyUI/output/experiments/ using ComfyUI's standard path helper
output_dir = folder_paths.get_output_directory()
base_path = os.path.join(output_dir, "experiments")
self.base_path = base_path
os.makedirs(self.base_path, exist_ok=True)
def save_run(self, image_tensor, metadata_json, baseline_image_tensor=None,
positive_prompt="", negative_prompt="", is_baseline=False):
"""
Save a complete experiment run to disk.
Args:
image_tensor: Output image from VAE decode (torch tensor or PIL Image)
metadata_json: JSON string containing FSampler metadata
baseline_image_tensor: Optional baseline image for comparison
positive_prompt: Positive prompt string
negative_prompt: Negative prompt string
is_baseline: Whether this run is a baseline (no skip mode)
Returns:
Path to the created experiment directory
"""
# Parse metadata
try:
metadata = json.loads(metadata_json) if isinstance(metadata_json, str) else metadata_json
except Exception as e:
print(f"[ExperimentLogger] Failed to parse metadata: {e}")
metadata = {}
# Create timestamped directory
timestamp_start = metadata.get("timestamp_start", time.time())
dt = datetime.fromtimestamp(timestamp_start)
timestamp_str = dt.strftime("%Y-%m-%d_%H-%M-%S")
unix_timestamp = int(timestamp_start)
# Extract model type(s)
model_type_str = ""
if "model_0" in metadata or "model_1" in metadata:
# Multi-model workflow
types = []
if "model_0" in metadata and metadata["model_0"].get("model_type"):
types.append(metadata["model_0"]["model_type"])
if "model_1" in metadata and metadata["model_1"].get("model_type"):
types.append(metadata["model_1"]["model_type"])
if types:
model_type_str = "_" + "+".join(types)
elif metadata.get("model_type"):
# Single model workflow
model_type_str = "_" + metadata["model_type"]
# Directory name: YYYY-MM-DD_HH-MM-SS_{unix_timestamp}_{model_type}
dir_name = f"{timestamp_str}_{unix_timestamp}{model_type_str}"
if is_baseline:
dir_name += "_baseline"
exp_dir = os.path.join(self.base_path, dir_name)
os.makedirs(exp_dir, exist_ok=True)
# Capture device info
device_info = self._get_device_info()
# Add device info and prompts to metadata
metadata["device_info"] = device_info
metadata["positive_prompt"] = positive_prompt
metadata["negative_prompt"] = negative_prompt
metadata["is_baseline"] = is_baseline
# Save output image
output_image_path = os.path.join(exp_dir, "output_image.png")
self._save_image(image_tensor, output_image_path)
# Save baseline image if provided
if baseline_image_tensor is not None:
baseline_image_path = os.path.join(exp_dir, "baseline.png")
self._save_image(baseline_image_tensor, baseline_image_path)
# Compute metrics (SSIM, RMSE, MAE) against baseline
try:
from .metrics import compute_metrics
metrics = compute_metrics(image_tensor, baseline_image_tensor)
metadata["metrics"] = metrics
print(f"[ExperimentLogger] Computed metrics: SSIM={metrics['ssim']:.4f}, RMSE={metrics['rmse']:.4f}, MAE={metrics['mae']:.4f}")
except Exception as e:
print(f"[ExperimentLogger] Failed to compute metrics: {e}")
metadata["metrics"] = None
# Save metadata JSON
metadata_path = os.path.join(exp_dir, "metadata.json")
with open(metadata_path, 'w') as f:
json.dump(metadata, f, indent=2)
# Save per-step data CSV
per_step_data = metadata.get("per_step_data", [])
if per_step_data:
csv_path = os.path.join(exp_dir, "per_step_data.csv")
self._save_per_step_csv(per_step_data, csv_path)
# Save sigmas CSV (if available in per_step_data)
if per_step_data:
sigmas_csv_path = os.path.join(exp_dir, "sigmas.csv")
self._save_sigmas_csv(per_step_data, sigmas_csv_path)
# Save summary.txt
summary_path = os.path.join(exp_dir, "summary.txt")
self._save_summary(metadata, summary_path)
print(f"[ExperimentLogger] Saved experiment to: {exp_dir}")
return exp_dir
def _get_device_info(self):
"""Capture device information (GPU, memory, etc.)."""
device_info = {}
try:
if torch.cuda.is_available():
device_info["device"] = f"cuda:{torch.cuda.current_device()}"
device_info["device_name"] = torch.cuda.get_device_name(torch.cuda.current_device())
device_info["total_memory_gb"] = torch.cuda.get_device_properties(0).total_memory / (1024**3)
device_info["allocated_memory_gb"] = torch.cuda.memory_allocated(0) / (1024**3)
device_info["reserved_memory_gb"] = torch.cuda.memory_reserved(0) / (1024**3)
else:
device_info["device"] = "cpu"
except Exception as e:
device_info["error"] = str(e)
return device_info
def _save_image(self, image_data, path):
"""
Save image tensor or PIL Image to disk.
Args:
image_data: Can be torch.Tensor (from VAE) or PIL.Image
path: Output path for PNG file
"""
try:
if isinstance(image_data, torch.Tensor):
# Convert tensor to PIL Image
# Assume tensor is in shape [B, C, H, W] or [C, H, W]
if image_data.dim() == 4:
# Batch dimension, take first image
image_data = image_data[0]
# Convert to numpy
img_np = image_data.cpu().numpy()
# If channels first, convert to channels last
if img_np.shape[0] in [1, 3, 4]:
img_np = np.transpose(img_np, (1, 2, 0))
# Clip and convert to uint8
img_np = np.clip(img_np * 255.0, 0, 255).astype(np.uint8)
# Convert to PIL
if img_np.shape[2] == 1:
img_pil = Image.fromarray(img_np[:, :, 0], mode='L')
elif img_np.shape[2] == 3:
img_pil = Image.fromarray(img_np, mode='RGB')
elif img_np.shape[2] == 4:
img_pil = Image.fromarray(img_np, mode='RGBA')
else:
raise ValueError(f"Unexpected number of channels: {img_np.shape[2]}")
img_pil.save(path)
elif isinstance(image_data, Image.Image):
# Already PIL Image
image_data.save(path)
else:
print(f"[ExperimentLogger] Unknown image type: {type(image_data)}")
except Exception as e:
print(f"[ExperimentLogger] Failed to save image to {path}: {e}")
def _save_per_step_csv(self, per_step_data, path):
"""Save per-step data to CSV."""
if not per_step_data:
return
try:
with open(path, 'w', newline='') as f:
# Get all unique keys from all step dicts
all_keys = set()
for step in per_step_data:
all_keys.update(step.keys())
fieldnames = sorted(all_keys)
writer = csv.DictWriter(f, fieldnames=fieldnames)
writer.writeheader()
writer.writerows(per_step_data)
except Exception as e:
print(f"[ExperimentLogger] Failed to save per-step CSV: {e}")
def _save_sigmas_csv(self, per_step_data, path):
"""Save sigma schedule to CSV."""
if not per_step_data:
return
try:
with open(path, 'w', newline='') as f:
writer = csv.writer(f)
writer.writerow(['step_index', 'sigma_current', 'sigma_next'])
for step in per_step_data:
writer.writerow([
step.get('step_index', ''),
step.get('sigma_current', ''),
step.get('sigma_next', '')
])
except Exception as e:
print(f"[ExperimentLogger] Failed to save sigmas CSV: {e}")
def _save_summary(self, metadata, path):
"""Save human-readable summary.txt."""
try:
with open(path, 'w') as f:
f.write("=" * 80 + "\n")
f.write("FSampler Experiment Run Summary\n")
f.write("=" * 80 + "\n\n")
# Check if this is a multi-model workflow
is_multi_model = "model_0" in metadata or "model_1" in metadata
if is_multi_model:
# Multi-model workflow
f.write("Multi-Model Workflow Detected\n")
f.write("=" * 80 + "\n\n")
for model_key in ["model_0", "model_1"]:
if model_key in metadata:
model_meta = metadata[model_key]
f.write(f"\n{'='*80}\n")
f.write(f"{model_key.upper()} (Base Model)\n" if model_key == "model_0" else f"{model_key.upper()} (Refiner Model)\n")
f.write(f"{'='*80}\n\n")
self._write_single_model_summary(f, model_meta)
# Global info at the bottom
f.write("\n" + "="*80 + "\n")
f.write("Global Information\n")
f.write("="*80 + "\n")
f.write(f"Is Baseline: {metadata.get('is_baseline', False)}\n")
# Device info
device_info = metadata.get("device_info", {})
if device_info:
f.write("\nDevice Information:\n")
f.write("-" * 80 + "\n")
for key, value in device_info.items():
f.write(f"{key}: {value}\n")
# Prompts
f.write("\nPrompts:\n")
f.write("-" * 80 + "\n")
f.write(f"Positive: {metadata.get('positive_prompt', 'N/A')}\n")
f.write(f"Negative: {metadata.get('negative_prompt', 'N/A')}\n")
# Metrics
if "metrics" in metadata and metadata["metrics"]:
f.write("\nImage Quality Metrics (vs Baseline):\n")
f.write("-" * 80 + "\n")
m = metadata["metrics"]
f.write(f"SSIM: {m['ssim']:.6f} (structural similarity, higher=better)\n")
f.write(f"RMSE: {m['rmse']:.6f} (root mean square error, lower=better)\n")
f.write(f"MAE: {m['mae']:.6f} (mean absolute error, lower=better)\n")
else:
# Single model workflow
self._write_single_model_summary(f, metadata)
# Device info
device_info = metadata.get("device_info", {})
if device_info:
f.write("Device Information:\n")
f.write("-" * 80 + "\n")
for key, value in device_info.items():
f.write(f"{key}: {value}\n")
f.write("\n")
# Prompts
f.write("Prompts:\n")
f.write("-" * 80 + "\n")
f.write(f"Positive: {metadata.get('positive_prompt', 'N/A')}\n")
f.write(f"Negative: {metadata.get('negative_prompt', 'N/A')}\n")
f.write("\n")
# Metrics
if "metrics" in metadata and metadata["metrics"]:
f.write("Image Quality Metrics (vs Baseline):\n")
f.write("-" * 80 + "\n")
m = metadata["metrics"]
f.write(f"SSIM: {m['ssim']:.6f} (structural similarity, higher=better)\n")
f.write(f"RMSE: {m['rmse']:.6f} (root mean square error, lower=better)\n")
f.write(f"MAE: {m['mae']:.6f} (mean absolute error, lower=better)\n")
f.write("\n")
f.write("=" * 80 + "\n")
except Exception as e:
print(f"[ExperimentLogger] Failed to save summary: {e}")
def _write_single_model_summary(self, f, metadata):
"""Write summary for a single model's metadata."""
# Basic info
f.write("Run Information:\n")
f.write("-" * 80 + "\n")
if "timestamp_start" in metadata:
dt = datetime.fromtimestamp(metadata["timestamp_start"])
f.write(f"Timestamp: {dt.strftime('%Y-%m-%d %H:%M:%S')}\n")
f.write(f"Model Type: {metadata.get('model_type', 'N/A')}\n")
f.write(f"Seed: {metadata.get('seed', 'N/A')}\n")
if "is_baseline" in metadata:
f.write(f"Is Baseline: {metadata.get('is_baseline', False)}\n")
f.write("\n")
# Sampler settings
f.write("Sampler Settings:\n")
f.write("-" * 80 + "\n")
f.write(f"Sampler: {metadata.get('sampler', 'N/A')}\n")
f.write(f"Scheduler: {metadata.get('scheduler', 'N/A')}\n")
f.write(f"Skip Mode: {metadata.get('skip_mode', 'N/A')}\n")
f.write(f"Adaptive Mode: {metadata.get('adaptive_mode', 'N/A')}\n")
f.write(f"Smoothing Beta: {metadata.get('smoothing_beta', 'N/A')}\n")
f.write(f"Protect First Steps: {metadata.get('protect_first_steps', 'N/A')}\n")
f.write(f"Protect Last Steps: {metadata.get('protect_last_steps', 'N/A')}\n")
if metadata.get('anchor_interval') is not None:
f.write(f"Anchor Interval: {metadata.get('anchor_interval')}\n")
if metadata.get('max_consecutive_skips') is not None:
f.write(f"Max Consecutive Skips: {metadata.get('max_consecutive_skips')}\n")
f.write("\n")
# Performance metrics
f.write("Performance Metrics:\n")
f.write("-" * 80 + "\n")
f.write(f"Total Steps: {metadata.get('total_steps', 'N/A')}\n")
f.write(f"Model Calls: {metadata.get('model_calls', 'N/A')}\n")
f.write(f"Skipped: {metadata.get('skipped', 'N/A')}\n")
f.write(f"Reduction: {metadata.get('reduction_percent', 0.0):.2f}%\n")
f.write(f"Total Time: {metadata.get('total_time_seconds', 0.0):.2f}s\n")
f.write("\n")
# Learning stabilizer stats
f.write("Learning Stabilizer (L) Statistics:\n")
f.write("-" * 80 + "\n")
f.write(f"L Final: {metadata.get('l_final', 1.0):.4f}\n")
f.write(f"L Mean: {metadata.get('l_mean', 1.0):.4f}\n")
f.write(f"L Min: {metadata.get('l_min', 1.0):.4f}\n")
f.write(f"L Max: {metadata.get('l_max', 1.0):.4f}\n")
f.write("\n")
|