From 7547e8ff4744339b92650b6ef3ff7405befe4beb Mon Sep 17 00:00:00 2001
From: skal <pascal.massimino@gmail.com>
Date: Thu, 12 Feb 2026 11:50:52 +0100
Subject: CNN v2: Patch-based training as default (like CNN v1)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Salient point detection on original images with patch extraction.

Changes:
- Added PatchDataset class (harris/fast/shi-tomasi/gradient detectors)
- Detects salient points on ORIGINAL images (no resize)
- Extracts 32×32 patches around salient points
- Default: 64 patches/image, harris detector
- Batch size: 16 (512 patches per batch)

Training modes:
1. Patch-based (default): --patch-size 32 --patches-per-image 64 --detector harris
2. Full-image (option): --full-image --image-size 256

Benefits:
- Focuses training on interesting regions
- Handles variable image sizes naturally
- Matches CNN v1 workflow
- Better convergence with limited data (8 images → 512 patches)

Script updated:
- train_cnn_v2_full.sh: Patch-based by default
- Configuration exposed for easy switching

Example:
  ./scripts/train_cnn_v2_full.sh  # Patch-based
  # Edit script: uncomment FULL_IMAGE for resize mode

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---
 scripts/train_cnn_v2_full.sh | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

(limited to 'scripts')

diff --git a/scripts/train_cnn_v2_full.sh b/scripts/train_cnn_v2_full.sh
index 119b788..4ddd9ac 100755
--- a/scripts/train_cnn_v2_full.sh
+++ b/scripts/train_cnn_v2_full.sh
@@ -14,8 +14,17 @@ CHECKPOINT_DIR="checkpoints"
 VALIDATION_DIR="validation_results"
 EPOCHS=10000
 CHECKPOINT_EVERY=500
-BATCH_SIZE=8
-IMAGE_SIZE=256
+BATCH_SIZE=16
+
+# Patch-based training (default)
+PATCH_SIZE=32
+PATCHES_PER_IMAGE=64
+DETECTOR="harris"
+
+# Full-image training (alternative - uncomment to use)
+# FULL_IMAGE="--full-image"
+# IMAGE_SIZE=256
+
 KERNEL_SIZES="1 3 5"
 CHANNELS="16 8 4"
 
@@ -31,13 +40,16 @@ echo "[1/4] Training CNN v2 model..."
 python3 training/train_cnn_v2.py \
   --input "$INPUT_DIR" \
   --target "$TARGET_DIR" \
-  --image-size $IMAGE_SIZE \
+  --patch-size $PATCH_SIZE \
+  --patches-per-image $PATCHES_PER_IMAGE \
+  --detector $DETECTOR \
   --kernel-sizes $KERNEL_SIZES \
   --channels $CHANNELS \
   --epochs $EPOCHS \
   --batch-size $BATCH_SIZE \
   --checkpoint-dir "$CHECKPOINT_DIR" \
-  --checkpoint-every $CHECKPOINT_EVERY
+  --checkpoint-every $CHECKPOINT_EVERY \
+  $FULL_IMAGE
 
 if [ $? -ne 0 ]; then
     echo "Error: Training failed"
-- 
cgit v1.2.3