code/scripts/plot_refuge2_loss.py

"""Plot REFUGE2 training loss curves from log file."""
import re
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import numpy as np

with open('/data/sichengli/Code/PixelGen/training_medical_refuge2.log', 'r') as f:
    text = f.read()

pattern = r'Epoch\s+(\d+):\s+100%.*?fm_loss=([\d.]+),\s*lpips_loss=([\d.]+),\s*loss=([\d.]+)'
matches = re.findall(pattern, text)

epochs, fm_losses, lpips_losses, total_losses = [], [], [], []
seen = set()
for m in matches:
    epoch = int(m[0])
    if epoch in seen:
        continue
    seen.add(epoch)
    epochs.append(epoch)
    fm_losses.append(float(m[1]))
    lpips_losses.append(float(m[2]))
    total_losses.append(float(m[3]))

steps = [e * 6 for e in epochs]
print(f"Total data points: {len(steps)}")
print(f"Step range: {steps[0]} - {steps[-1]}")

def smooth(values, window=50):
    if len(values) < window:
        return values
    kernel = np.ones(window) / window
    return np.convolve(values, kernel, mode='valid')

fig, axes = plt.subplots(1, 3, figsize=(18, 5))
for ax, data, label, color in zip(
    axes,
    [fm_losses, lpips_losses, total_losses],
    ['FM Loss (MSE)', 'LPIPS Loss', 'Total Loss'],
    ['#2196F3', '#FF9800', '#4CAF50']
):
    s = np.array(steps)
    d = np.array(data)
    ax.plot(s, d, alpha=0.15, color=color, linewidth=0.5)
    w = min(100, max(1, len(d) // 5))
    if w > 1:
        d_smooth = smooth(d, w)
        s_smooth = s[w-1:][:len(d_smooth)]
        ax.plot(s_smooth, d_smooth, color=color, linewidth=2, label=label + " (smoothed)")
    ax.set_xlabel('Training Steps', fontsize=12)
    ax.set_ylabel('Loss', fontsize=12)
    ax.set_title(label, fontsize=14, fontweight='bold')
    ax.legend(fontsize=10)
    ax.grid(True, alpha=0.3)
    ax.set_xlim(0, 100000)

plt.suptitle('REFUGE2 Training Loss Curves (100k steps)', fontsize=16, fontweight='bold', y=1.02)
plt.tight_layout()
out_path = '/data/sichengli/Code/PixelGen/medical_workdirs/exp_PixelGen_Medical_REFUGE2/val_samples/loss_curves.png'
plt.savefig(out_path, dpi=150, bbox_inches='tight')
print(f"Saved: {out_path}")

milestones = [0, 10000, 20000, 30000, 50000, 70000, 100000]
header = f"{'Step':>8s} | {'FM Loss':>10s} | {'LPIPS':>10s} | {'Total':>10s}"
print(f"\n{header}")
print('-' * len(header))
for ms in milestones:
    idx = min(range(len(steps)), key=lambda i: abs(steps[i] - ms))
    print(f"{steps[idx]:>8d} | {fm_losses[idx]:>10.4f} | {lpips_losses[idx]:>10.4f} | {total_losses[idx]:>10.4f}")