|
|
|
|
| import numpy as np |
| import matplotlib.pyplot as plt |
| import torch |
|
|
|
|
| def to_numpy_img(t): |
| """Convert torch CHW / numpy HWC (or HW) to an array matplotlib can show.""" |
| if isinstance(t, torch.Tensor): |
| t = t.detach().cpu().numpy() |
| if t.ndim == 4: |
| t = t[0] |
| if t.ndim == 3: |
| t = np.transpose(t, (1, 2, 0)) |
| elif isinstance(t, np.ndarray) and t.ndim == 3: |
| |
| if t.shape[0] in (1, 3, 4) and t.shape[-1] not in (1, 3, 4): |
| t = np.transpose(t, (1, 2, 0)) |
|
|
| if isinstance(t, np.ndarray) and t.ndim == 3 and t.shape[-1] == 1: |
| t = t.squeeze(-1) |
| return t |
|
|
|
|
| def show_synthesis_result(normal_img, jnd_map, anomaly_img, severity_mask, panel_name: str = ""): |
|
|
| fig, axes = plt.subplots(1, 4, figsize=(16, 4)) |
| axes[0].imshow(to_numpy_img(normal_img)) |
| axes[0].set_title("Normal Image") |
| axes[1].imshow(to_numpy_img(jnd_map), cmap="viridis") |
| axes[1].set_title("JND Map") |
| axes[2].imshow(to_numpy_img(anomaly_img)) |
| axes[2].set_title(f"Synthesized Anomaly (panel={panel_name})") |
| axes[3].imshow(to_numpy_img(severity_mask), cmap="jet", vmin=0, vmax=3) |
| axes[3].set_title("Severity Mask (0=bg,1=easy,2=med,3=hard)") |
| for ax in axes: |
| ax.axis("off") |
| plt.tight_layout() |
| plt.show() |
|
|
|
|
| def show_model_output(anomaly_img, rec_img, rec_jnd, seg_map, gt_mask=None): |
|
|
| n_cols = 5 if gt_mask is not None else 4 |
| fig, axes = plt.subplots(1, n_cols, figsize=(4 * n_cols, 4)) |
| axes[0].imshow(to_numpy_img(anomaly_img)) |
| axes[0].set_title("Input (anomaly)") |
| axes[1].imshow(to_numpy_img(rec_img)) |
| axes[1].set_title("Reconstruction") |
| axes[2].imshow(to_numpy_img(rec_jnd), cmap="viridis") |
| axes[2].set_title("Reconstructed JND") |
| axes[3].imshow(to_numpy_img(seg_map), cmap="jet") |
| axes[3].set_title("Predicted anomaly map") |
| if gt_mask is not None: |
| axes[4].imshow(to_numpy_img(gt_mask), cmap="gray") |
| axes[4].set_title("Ground Truth mask") |
| for ax in axes: |
| ax.axis("off") |
| plt.tight_layout() |
| plt.show() |
|
|
|
|
| def plot_loss_curves(history: dict): |
|
|
| plt.figure(figsize=(8, 5)) |
| for name, values in history.items(): |
| plt.plot(values, label=name) |
| plt.xlabel("Epoch") |
| plt.ylabel("Loss") |
| plt.legend() |
| plt.title("Training Loss Curves") |
| plt.grid(alpha=0.3) |
| plt.show() |
|
|
