added spectral LRP analysis implementation

visualize_lrp.py

@@ -27,8 +27,6 @@
                                 "/prodi/hpcmem/spots_ftir/CO1801a/",
                                 "/prodi/hpcmem/spots_ftir/CO722/"]
 
-DATA_OUT: Final[str] = "/prodi/hpcmem/spots_ftir/data_out/"
-
 MODEL_OUTPUT_PATH: Final[str] = "./models/"
 
 
@@ -67,7 +65,7 @@ def main() -> None:
 
     # load a single image to get the input shape
     # train data has the shape (batch_size, channels, width, height) -> (BATCH_SIZE, 442, 400, 400)
-    print("[ creating model ]")
+    print("[ creating models ]")
     tmp_data, _ = next(iter(test_loader))
     in_channels = tmp_data.shape[1]
 
@@ -137,8 +135,27 @@ def main() -> None:
         attribution_depth = lrp_model_depth.attribute(img, target=None).cpu().detach().numpy()[0]
 
         img = img.cpu().detach().numpy()[0]
+        input_spectral = np.copy(img)
         img = np.mean(img, axis=0)
 
+        # specral data
+        input_spectral = input_spectral.mean(axis=(1, 2))
+        attribution_spectral_normal = attribution_normal.mean(axis=(1, 2))
+        attribution_spectral_depth = attribution_depth.mean(axis=(1, 2))
+
+        # normalize the spectra data
+        peak_interval = input_spectral[339:379]
+        peak_point = np.max(peak_interval)
+        input_spectral = input_spectral / peak_point
+
+        peak_interval = attribution_spectral_normal[339:379]
+        peak_point = np.max(peak_interval)
+        attribution_spectral_normal = attribution_spectral_normal / peak_point
+
+        peak_interval = attribution_spectral_depth[339:379]
+        peak_point = np.max(peak_interval)
+        attribution_spectral_depth = attribution_spectral_depth / peak_point
+
         # add the attributions of every channel dimension up
         attribution_normal = np.sum(attribution_normal, axis=0)
         attribution_normal = np.expand_dims(attribution_normal, axis=0)
@@ -155,15 +172,13 @@ def main() -> None:
         cmap = LinearSegmentedColormap.from_list(
             "RdWhGn", ["red", "white", "green"]
         )
+        norm = mpl.colors.Normalize(vmin=-1, vmax=1)
 
+        # create and save the normal LRP analysis
         fig, axes = plt.subplots(1, 3, figsize=(10, 3))
-        #axis_separator = make_axes_locatable(axes[2])
-        #colorbar_axis = axis_separator.append_axes("bottom", size="5%", pad=0.1)
-        norm = mpl.colors.Normalize(vmin=-1, vmax=1)
         plt.colorbar(mpl.cm.ScalarMappable(norm=norm, cmap=cmap),
                      orientation="vertical", ax=axes,
                      label="Attribution Values")
-
         axes[0].grid()
         axes[0].imshow(img)
         axes[0].set_title("Input Image")
@@ -183,6 +198,26 @@ def main() -> None:
         fig.savefig(f"./plots/{model_file}_lrp.png")
         plt.close()
 
+        # create and save the LRP analysis for the specral channels
+        fig, axes = plt.subplots(1, 4, figsize=(20, 5))
+        axes[0].grid()
+        axes[0].imshow(img)
+        axes[0].set_title("Input Image")
+        axes[0].set_xticks([])
+        axes[0].set_yticks([])
+        axes[1].grid()
+        axes[1].plot(input_spectral)
+        axes[1].set_title("Spectral Input")
+        axes[2].grid()
+        axes[2].plot(attribution_spectral_normal)
+        axes[2].set_title("Spectral Attribution Normal")
+        axes[3].grid()
+        axes[3].plot(attribution_spectral_depth)
+        axes[3].set_title("Spectral Attribution Depthwise")
+
+        fig.savefig(f"./plots/{model_file}_spectral_lrp.png")
+        plt.close()
+
     print("[ finished LRP analysis ]")
 
 if __name__ == "__main__":