7_inpainting.sh

#/bin/bash
#
# Anh Nguyen <anh.ng8@gmail.com>
# 2016

# Take in an unit number
if [ "$#" -ne "1" ]; then
  echo "Provide 1 output unit number e.g. 945 for bell pepper."
  exit 1
fi

opt_layer=fc6
act_layer=fc8
units="${1}"      # Index of neurons in fc layers or channels in conv layers
xy=0              # Spatial position for conv layers, for fc layers: xy = 0

# Note: these are the same params as used for "inpainting" experiments in the paper; 
# however, careful tuning may yield better results

n_iters=20        # Run for N iterations
reset_every=0     # Reset the code every N iterations (for diversity)
save_every=0      # Save a sample every N iterations
lr=1 
lr_end=1e-32      # Linearly decay toward this ending lr (e.g. for decaying toward 0, set lr_end = 1e-10)
threshold=0       # Filter out samples below this threshold e.g. 0.98

# -----------------------------------------------
# Multipliers in the update rule Eq.11 in the paper
# -----------------------------------------------
epsilon1=1e-5      # prior
epsilon2=1         # condition
epsilon3=0         # noise
epsilon4=1e-9
# -----------------------------------------------

init_file="images/volcano.jpg"    # Start from a random code. To start from a real code, replace with a path e.g. "images/filename.jpg"

# Condition net
net_weights="nets/caffenet/bvlc_reference_caffenet.caffemodel"
net_definition="nets/caffenet/caffenet.prototxt"
#-----------------------

# Output dir
output_dir="output/${act_layer}_chain_${units}_eps1_${epsilon1}_eps3_${epsilon3}"
mkdir -p ${output_dir}

# Directory to store samples
if [ "${save_every}" -gt "0" ]; then
    sample_dir=${output_dir}/samples
    rm -rf ${sample_dir} 
    mkdir -p ${sample_dir} 
fi

for unit in ${units}; do
    unit_pad=`printf "%04d" ${unit}`

    # For in-painting experiments, a different seed will result in a new random position for the masking patch
    for seed in {0..9}; do

        python ./sampling_class.py \
            --act_layer ${act_layer} \
            --opt_layer ${opt_layer} \
            --unit ${unit} \
            --xy ${xy} \
            --n_iters ${n_iters} \
            --save_every ${save_every} \
            --reset_every ${reset_every} \
            --lr ${lr} \
            --lr_end ${lr_end} \
            --seed ${seed} \
            --output_dir ${output_dir} \
            --init_file ${init_file} \
            --epsilon1 ${epsilon1} \
            --epsilon2 ${epsilon2} \
            --epsilon3 ${epsilon3} \
            --epsilon4 ${epsilon4} \
            --threshold ${threshold} \
            --net_weights ${net_weights} \
            --net_definition ${net_definition} \

    done

    # Make a montage of steps
    f=${output_dir}/inpainting.jpg
    montage ${output_dir}/${act_layer}_*.jpg -tile 5x -geometry +1+1 ${f}
    readlink -f ${f}

done