diff --git a/datasets/raise/attribute_source_camera.py b/datasets/raise/attribute_source_camera.py
index c03a339..ad6b46e 100755
--- a/datasets/raise/attribute_source_camera.py
+++ b/datasets/raise/attribute_source_camera.py
@@ -102,6 +102,17 @@ from utils import silentTqdm
 
 returnSingleColorChannelImage = lambda singleColorChannelImage, _minColor, _maxColor: singleColorChannelImage
 
+# Assume to have `{min,max}Color` hardcoded.
+print('Load training images to memory')
+rescaleIfNeeded = rescaleRawImageForDenoiser
+cameraTrainingImages = {}
+for cameraTrainingImageIndex in tqdm(range(numberOfTrainingImages), 'Camera training image index'):
+    for cameraIndex, camera in enumerate(tqdm(IMAGES_CAMERAS_FOLDER, 'Camera')):
+        singleColorChannelImages = getSingleColorChannelImages(camera, cameraTrainingImageIndex)
+        multipleColorsImage = getMultipleColorsImage(singleColorChannelImages)
+        cameraTrainingImages[camera] = cameraTrainingImages.get(camera, []) + [multipleColorsImage]
+print('Training images loaded to memory')
+
 # 2 loops:
 # - the first one is about computing `{min,max}Color`
 # - the second one is about estimating better and better the PRNU of each camera, as consider more and more training images and measuring the resulting attribution of cameras
@@ -143,11 +154,7 @@ for computeExtremes in tqdm(([True] if minColor is None or maxColor is None else
                 cameraIterativeMean.add(imagePrnuEstimateNpArray)
             else:
                 # Still use `cameraIterativeMean` to simplify the implementation.
-                cameraIterativeMean.mean = np.mean([cameraTrainingImage - mergeSingleColorChannelImagesAccordingToBayerFilter({color: cameraColorMeans[camera][color].mean for color in Color}) for cameraTrainingImage in [multipleColorsImage]], axis = 0)#[:cameraTrainingImageIndex + 1]])
-                print(f'{cameraIterativeMean.mean = }')
-                print(f'{camerasIterativeMean[camera].mean = }')
-                if cameraIterativeMean.mean[0, 0] != 0:
-                    exit(1)
+                cameraIterativeMean.mean = np.mean([cameraTrainingImage - mergeSingleColorChannelImagesAccordingToBayerFilter({color: cameraColorMeans[camera][color].mean for color in Color}) for cameraTrainingImage in cameraTrainingImages[camera][:cameraTrainingImageIndex + 1]], axis = 0)
             # If we are considering the last camera and (not `PREDICT_ONLY_ON_WHOLE_TRAINING_SET` or we are considering the last training image), then we proceeded an additional image for all cameras and we can predict the accuracy at this learning step.
             if cameraIndex == numberOfCameras - 1 and (not PREDICT_ONLY_ON_WHOLE_TRAINING_SET or cameraTrainingImageIndex == numberOfTrainingImages - 1):
                 numberOfTrainingImagesAccuracy = 0