#!/bin/bash
#
# Test the hybrid text+CLIP matching approach for logo detection.
#
# This approach uses text recognition to improve logo matching:
# - If reference logo has text and detection matches it: use lower CLIP threshold
# - If reference logo has text but detection doesn't match: use higher CLIP threshold
# - If reference logo has no text: use standard CLIP threshold
#
# Usage:
#   ./run_hybrid_test.sh
#

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
OUTPUT_FILE="${SCRIPT_DIR}/test_results/hybrid_matching_results.txt"

# Model - baseline CLIP
MODEL="openai/clip-vit-large-patch14"

# Fixed parameters
NUM_LOGOS=20
REFS_PER_LOGO=10
POSITIVE_SAMPLES=20
NEGATIVE_SAMPLES=100
SEED=42

# Create output directory if needed
mkdir -p "${SCRIPT_DIR}/test_results"

# Clear output file and write header
cat > "$OUTPUT_FILE" << EOF
Hybrid Text+CLIP Matching Test Results
======================================
Date: $(date)

Model: ${MODEL}

Fixed Parameters:
  Number of logo brands: ${NUM_LOGOS}
  Refs per logo: ${REFS_PER_LOGO}
  Positive samples/logo: ${POSITIVE_SAMPLES}
  Negative samples/logo: ${NEGATIVE_SAMPLES}
  Seed: ${SEED}

EOF

echo "Hybrid Text+CLIP Matching Test"
echo "==============================="
echo "Model: ${MODEL}"
echo ""

# Test 1: Compare hybrid vs multi-ref baseline
echo "=== Test 1: Multi-ref baseline (for comparison) ==="
echo "" >> "$OUTPUT_FILE"
echo "=== BASELINE: Multi-ref (max) at threshold 0.70 ===" >> "$OUTPUT_FILE"

uv run python "$SCRIPT_DIR/test_logo_detection.py" \
    --num-logos $NUM_LOGOS \
    --refs-per-logo $REFS_PER_LOGO \
    --positive-samples $POSITIVE_SAMPLES \
    --negative-samples $NEGATIVE_SAMPLES \
    --matching-method multi-ref \
    --min-matching-refs 1 \
    --use-max-similarity \
    --threshold 0.70 \
    --margin 0.05 \
    --seed $SEED \
    --embedding-model "$MODEL" \
    --output-file "$OUTPUT_FILE" \
    --no-cache

echo ""

# Test 2: Hybrid with default thresholds
echo "=== Test 2: Hybrid with default thresholds ==="
echo "" >> "$OUTPUT_FILE"
echo "=== HYBRID: default thresholds (0.70/0.60/0.80) ===" >> "$OUTPUT_FILE"

uv run python "$SCRIPT_DIR/test_logo_detection.py" \
    --num-logos $NUM_LOGOS \
    --refs-per-logo $REFS_PER_LOGO \
    --positive-samples $POSITIVE_SAMPLES \
    --negative-samples $NEGATIVE_SAMPLES \
    --matching-method hybrid \
    --threshold 0.70 \
    --hybrid-text-threshold 0.60 \
    --hybrid-no-text-threshold 0.80 \
    --text-similarity-threshold 0.5 \
    --margin 0.05 \
    --seed $SEED \
    --embedding-model "$MODEL" \
    --output-file "$OUTPUT_FILE" \
    --no-cache

echo ""

# Test 3: Hybrid with more aggressive text bonus
echo "=== Test 3: Hybrid with lower text-match threshold ==="
echo "" >> "$OUTPUT_FILE"
echo "=== HYBRID: aggressive text bonus (0.70/0.55/0.80) ===" >> "$OUTPUT_FILE"

uv run python "$SCRIPT_DIR/test_logo_detection.py" \
    --num-logos $NUM_LOGOS \
    --refs-per-logo $REFS_PER_LOGO \
    --positive-samples $POSITIVE_SAMPLES \
    --negative-samples $NEGATIVE_SAMPLES \
    --matching-method hybrid \
    --threshold 0.70 \
    --hybrid-text-threshold 0.55 \
    --hybrid-no-text-threshold 0.80 \
    --text-similarity-threshold 0.5 \
    --margin 0.05 \
    --seed $SEED \
    --embedding-model "$MODEL" \
    --output-file "$OUTPUT_FILE" \
    --no-cache

echo ""

# Test 4: Hybrid with stricter text mismatch penalty
echo "=== Test 4: Hybrid with stricter text mismatch penalty ==="
echo "" >> "$OUTPUT_FILE"
echo "=== HYBRID: strict mismatch (0.70/0.60/0.85) ===" >> "$OUTPUT_FILE"

uv run python "$SCRIPT_DIR/test_logo_detection.py" \
    --num-logos $NUM_LOGOS \
    --refs-per-logo $REFS_PER_LOGO \
    --positive-samples $POSITIVE_SAMPLES \
    --negative-samples $NEGATIVE_SAMPLES \
    --matching-method hybrid \
    --threshold 0.70 \
    --hybrid-text-threshold 0.60 \
    --hybrid-no-text-threshold 0.85 \
    --text-similarity-threshold 0.5 \
    --margin 0.05 \
    --seed $SEED \
    --embedding-model "$MODEL" \
    --output-file "$OUTPUT_FILE" \
    --no-cache

echo ""

# Test 5: Hybrid with lower text similarity threshold (more lenient OCR matching)
echo "=== Test 5: Hybrid with lenient text matching ==="
echo "" >> "$OUTPUT_FILE"
echo "=== HYBRID: lenient text matching (text_sim=0.4) ===" >> "$OUTPUT_FILE"

uv run python "$SCRIPT_DIR/test_logo_detection.py" \
    --num-logos $NUM_LOGOS \
    --refs-per-logo $REFS_PER_LOGO \
    --positive-samples $POSITIVE_SAMPLES \
    --negative-samples $NEGATIVE_SAMPLES \
    --matching-method hybrid \
    --threshold 0.70 \
    --hybrid-text-threshold 0.60 \
    --hybrid-no-text-threshold 0.80 \
    --text-similarity-threshold 0.4 \
    --margin 0.05 \
    --seed $SEED \
    --embedding-model "$MODEL" \
    --output-file "$OUTPUT_FILE" \
    --no-cache

echo ""
echo "======================================="
echo "Tests complete!"
echo "Results saved to: $OUTPUT_FILE"
echo "======================================="