File size: 10,291 Bytes
f16e7e0 | 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 | # utils.py — shared helpers used across all modules
import os
import yaml
import logging
import numpy as np
import matplotlib.pyplot as plt
import cv2
import tensorflow as tf
from tensorflow.keras.preprocessing.image import load_img
# ---------------------------------------------------------------------------
# Logger
# ---------------------------------------------------------------------------
def get_logger(name: str, log_dir: str = "./logs") -> logging.Logger:
os.makedirs(log_dir, exist_ok=True)
logger = logging.getLogger(name)
if logger.handlers:
return logger
logger.setLevel(logging.INFO)
fmt = logging.Formatter(
"%(asctime)s | %(levelname)s | %(name)s | %(message)s",
datefmt="%Y-%m-%d %H:%M:%S"
)
ch = logging.StreamHandler()
ch.setFormatter(fmt)
logger.addHandler(ch)
fh = logging.FileHandler(
os.path.join(log_dir, f"{name}.log"), encoding="utf-8"
)
fh.setFormatter(fmt)
logger.addHandler(fh)
return logger
# ---------------------------------------------------------------------------
# Config loader
# ---------------------------------------------------------------------------
def load_config(path: str = "config.yaml") -> dict:
with open(path) as f:
return yaml.safe_load(f)
# ---------------------------------------------------------------------------
# Plotting
# ---------------------------------------------------------------------------
def plot_history(history, title: str, save_path: str = None):
fig, axes = plt.subplots(1, 2, figsize=(13, 4))
axes[0].plot(history.history["accuracy"], label="Train")
axes[0].plot(history.history["val_accuracy"], label="Val")
axes[0].set_title(f"{title} - Accuracy")
axes[0].set_xlabel("Epoch")
axes[0].legend()
axes[1].plot(history.history["loss"], label="Train")
axes[1].plot(history.history["val_loss"], label="Val")
axes[1].set_title(f"{title} - Loss")
axes[1].set_xlabel("Epoch")
axes[1].legend()
plt.tight_layout()
if save_path:
plt.savefig(save_path, bbox_inches="tight", dpi=100)
plt.show()
plt.close()
def plot_comparison(results: dict, save_path: str = None):
plt.figure(figsize=(10, 4))
colors = [
"crimson" if v == max(results.values()) else "steelblue"
for v in results.values()
]
bars = plt.bar(results.keys(), results.values(), color=colors)
plt.bar_label(bars, fmt="%.4f", padding=3)
plt.ylim(min(results.values()) - 0.05, 1.0)
plt.title("Model Comparison - Validation Accuracy (red = best)")
plt.ylabel("Val Accuracy")
plt.xticks(rotation=15)
plt.tight_layout()
if save_path:
plt.savefig(save_path, bbox_inches="tight", dpi=100)
plt.show()
plt.close()
# ---------------------------------------------------------------------------
# Grad-CAM helpers
# ---------------------------------------------------------------------------
def _collect_all_layers(model) -> list:
"""
Flatten all layers from a model including layers inside nested sub-models.
Returns a flat list of (layer_object, parent_model) tuples.
"""
result = []
def _recurse(m):
for layer in m.layers:
result.append(layer)
if hasattr(layer, "layers") and len(layer.layers) > 0:
_recurse(layer)
_recurse(model)
return result
def get_last_conv_layer(model) -> str:
"""
Return the name of the last Conv2D layer found anywhere inside the model,
including inside nested sub-models (MobileNetV2, EfficientNetB0 etc.).
"""
all_layers = _collect_all_layers(model)
conv_layers = [l for l in all_layers if isinstance(l, tf.keras.layers.Conv2D)]
if not conv_layers:
raise ValueError("No Conv2D layer found in model.")
return conv_layers[-1].name
def _build_gradcam_model(model, last_conv_layer_name: str):
"""
Build a Grad-CAM sub-model that outputs:
[conv_layer_output, final_model_predictions]
Works for both:
- Plain CNNs: Conv2D layers are direct children of the model
- Nested models: Conv2D is inside a sub-model (MobileNetV2, EfficientNetB0)
Strategy: find which sub-model owns the target conv layer, build a
feature extractor from that sub-model's input to [conv_output, sub_output],
then chain it with the remaining head layers of the outer model.
"""
all_layers = _collect_all_layers(model)
# Find the layer object
target_layer = None
for layer in all_layers:
if layer.name == last_conv_layer_name:
target_layer = layer
break
if target_layer is None:
raise ValueError(f"Layer '{last_conv_layer_name}' not found in model.")
# Check if the conv layer is a direct child of the outer model
direct_names = [l.name for l in model.layers]
if last_conv_layer_name in direct_names:
# Plain CNN — simple case
grad_model = tf.keras.models.Model(
inputs = model.input,
outputs = [model.get_layer(last_conv_layer_name).output, model.output]
)
return grad_model, None # None = no separate head needed
# Nested model case — find which direct child sub-model contains the layer
owner_submodel = None
for layer in model.layers:
if hasattr(layer, "layers"):
sub_names = [l.name for l in _collect_all_layers(layer)]
if last_conv_layer_name in sub_names:
owner_submodel = layer
break
if owner_submodel is None:
raise ValueError(
f"Could not find parent sub-model for layer '{last_conv_layer_name}'."
)
# Build: sub-model input -> [conv_output, sub_model_output]
sub_grad_model = tf.keras.models.Model(
inputs = owner_submodel.input,
outputs = [
owner_submodel.get_layer(last_conv_layer_name).output,
owner_submodel.output,
]
)
# Collect head layers (everything after the sub-model in the outer model)
head_layers = []
found = False
for layer in model.layers:
if found:
head_layers.append(layer)
if layer.name == owner_submodel.name:
found = True
return sub_grad_model, head_layers
def get_gradcam_heatmap(model, img_array: np.ndarray, last_conv_layer_name: str):
"""
Compute Grad-CAM heatmap.
Parameters
----------
model : compiled Keras model
img_array : preprocessed image, shape (1, H, W, 3), values in [0,1]
last_conv_layer_name: name of the target Conv2D layer
Returns
-------
heatmap : np.ndarray shape (H_conv, W_conv), values in [0,1]
pred_idx : int, predicted class index
"""
grad_model, head_layers = _build_gradcam_model(model, last_conv_layer_name)
with tf.GradientTape() as tape:
if head_layers is None:
# Plain CNN — single forward pass
conv_outputs, predictions = grad_model(img_array)
else:
# Nested model — two-stage forward pass
conv_outputs, sub_output = grad_model(img_array)
# Run through head layers sequentially
x = sub_output
for layer in head_layers:
x = layer(x)
predictions = x
pred_idx = tf.argmax(predictions[0])
loss = predictions[:, pred_idx]
# Watch conv_outputs so we can compute gradients w.r.t. it
tape.watch(conv_outputs)
# Recompute with watched tensor inside tape scope
with tf.GradientTape() as tape2:
tape2.watch(conv_outputs)
if head_layers is None:
conv_out_val, preds = grad_model(img_array)
else:
conv_out_val, sub_out = grad_model(img_array)
x = sub_out
for layer in head_layers:
x = layer(x)
preds = x
pred_idx = int(tf.argmax(preds[0]))
class_loss = preds[:, pred_idx]
grads = tape2.gradient(class_loss, conv_out_val)
if grads is None:
raise ValueError(
"Gradients are None. The conv layer output is not part of the "
"computation graph. Try a different layer name."
)
pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
heatmap = conv_out_val[0] @ pooled_grads[..., tf.newaxis]
heatmap = tf.squeeze(heatmap)
heatmap = tf.maximum(heatmap, 0)
heatmap = heatmap / (tf.math.reduce_max(heatmap) + 1e-8)
return heatmap.numpy(), pred_idx
# ---------------------------------------------------------------------------
# Full Grad-CAM visualisation
# ---------------------------------------------------------------------------
def generate_gradcam_overlay(model, img_path: str, last_conv_layer: str,
image_size: tuple, class_names: list,
save_path: str = None):
img = load_img(img_path, target_size=image_size)
img_array = np.array(img) / 255.0
img_input = np.expand_dims(img_array, axis=0).astype(np.float32)
heatmap, pred_idx = get_gradcam_heatmap(model, img_input, last_conv_layer)
heatmap_resized = cv2.resize(heatmap, image_size)
heatmap_colored = cv2.cvtColor(
cv2.applyColorMap(np.uint8(255 * heatmap_resized), cv2.COLORMAP_JET),
cv2.COLOR_BGR2RGB
)
overlay = cv2.addWeighted(
np.uint8(255 * img_array), 0.6, heatmap_colored, 0.4, 0
)
probs = model.predict(img_input, verbose=0)[0]
conf = probs[pred_idx] * 100
fig, axes = plt.subplots(1, 3, figsize=(13, 4))
axes[0].imshow(img)
axes[0].set_title("Original MRI")
axes[0].axis("off")
axes[1].imshow(heatmap_resized, cmap="jet")
axes[1].set_title("Grad-CAM Heatmap")
axes[1].axis("off")
axes[2].imshow(overlay)
axes[2].set_title(f"Pred: {class_names[pred_idx]} ({conf:.1f}%)")
axes[2].axis("off")
plt.suptitle(
f"Grad-CAM - {class_names[pred_idx].upper()}",
fontsize=14, fontweight="bold"
)
plt.tight_layout()
if save_path:
plt.savefig(save_path, bbox_inches="tight", dpi=100)
plt.show()
plt.close()
return pred_idx, conf, overlay |