CNN v2 training: Add --grayscale-loss option for luminance-based loss computation

Add option to compute loss on grayscale (Y = 0.299*R + 0.587*G + 0.114*B) instead of full RGBA channels. Useful for training models that prioritize luminance accuracy over color accuracy.

Changes:
- training/train_cnn_v2.py: Add --grayscale-loss flag and grayscale conversion in loss computation
- scripts/train_cnn_v2_full.sh: Add --grayscale-loss parameter support
- doc/CNN_V2.md: Document grayscale loss in training configuration and checkpoint format
- doc/HOWTO.md: Add usage examples for --grayscale-loss flag

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

1 files changed, 15 insertions, 1 deletions

diff --git a/scripts/train_cnn_v2_full.sh b/scripts/train_cnn_v2_full.sh
index 8b09191..e444f20 100755
--- a/scripts/train_cnn_v2_full.sh
+++ b/scripts/train_cnn_v2_full.sh
@@ -16,6 +16,7 @@
 #   --kernel-sizes K          Comma-separated kernel sizes (default: 3,3,3)
 #   --num-layers N            Number of layers (default: 3)
 #   --mip-level N             Mip level for p0-p3 features: 0-3 (default: 0)
+#   --grayscale-loss          Compute loss on grayscale instead of RGBA
 #
 # PATCH PARAMETERS:
 #   --patch-size N            Patch size (default: 8)
@@ -60,6 +61,7 @@ DETECTOR="harris"
 KERNEL_SIZES="3,3,3"
 NUM_LAYERS=3
 MIP_LEVEL=0
+GRAYSCALE_LOSS=false
 FULL_IMAGE_MODE=false
 IMAGE_SIZE=256
 
@@ -143,6 +145,10 @@ while [[ $# -gt 0 ]]; do
             MIP_LEVEL="$2"
             shift 2
             ;;
+        --grayscale-loss)
+            GRAYSCALE_LOSS=true
+            shift
+            ;;
         --patch-size)
             if [ -z "$2" ]; then
                 echo "Error: --patch-size requires a number argument"
@@ -267,6 +273,13 @@ fi
 if [ "$VALIDATE_ONLY" = false ]; then
     # Step 1: Train model
     echo "[1/4] Training CNN v2 model..."
+
+# Build optional flags
+OPTIONAL_FLAGS=""
+if [ "$GRAYSCALE_LOSS" = true ]; then
+    OPTIONAL_FLAGS="$OPTIONAL_FLAGS --grayscale-loss"
+fi
+
 python3 training/train_cnn_v2.py \
   --input "$INPUT_DIR" \
   --target "$TARGET_DIR" \
@@ -277,7 +290,8 @@ python3 training/train_cnn_v2.py \
   --epochs $EPOCHS \
   --batch-size $BATCH_SIZE \
   --checkpoint-dir "$CHECKPOINT_DIR" \
-  --checkpoint-every $CHECKPOINT_EVERY
+  --checkpoint-every $CHECKPOINT_EVERY \
+  $OPTIONAL_FLAGS
 
 if [ $? -ne 0 ]; then
     echo "Error: Training failed"