advanced-semantic-attention-topk-2lexical-eval.sh

#!/bin/bash

DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"

model_name="show_and_tell_advanced_model_topk_semantic_attention_2lexical"
num_processes=1
gpu_fraction=0.97
device=0
model=ShowAndTellAdvancedModel

MODEL_DIR="${DIR}/model/${model_name}"
for ckpt in $(ls ${MODEL_DIR} | python ${DIR}/tools/every_n_step.py 20000); do 
  # the script directory
  VALIDATE_IMAGE_DIR="${DIR}/data/ai_challenger_caption_validation_20170910/caption_validation_images_20170910"
  VALIDATE_REFERENCE_FILE="${DIR}/data/ai_challenger_caption_validation_20170910/reference.json"

  CHECKPOINT_PATH="${MODEL_DIR}/model.ckpt-$ckpt"
  OUTPUT_DIR="${MODEL_DIR}/model.ckpt-${ckpt}.eval"

  mkdir $OUTPUT_DIR

  cd ${DIR}/im2txt

  for prefix in 0 1 2 3 4 5 6 7 8 9 a b c d e f; do 
    if [ ! -f ${OUTPUT_DIR}/part-${prefix}.json ]; then
      echo "CUDA_VISIBLE_DEVICES=$device python inference.py \
        --input_file_pattern='${VALIDATE_IMAGE_DIR}/${prefix}*.jpg' \
        --checkpoint_path=${CHECKPOINT_PATH} \
        --vocab_file=${DIR}/data/word_counts.txt \
        --output=${OUTPUT_DIR}/part-${prefix}.json \
        --model=${model} \
        --attention_mechanism=BahdanauAttention \
        --num_lstm_layers=1 \
        --predict_words_via_image_output=True \
        --use_semantic_attention=True \
        --use_separate_embedding_for_semantic_attention=True \
        --semantic_attention_type="topk" \
        --use_lexical_embedding=True \
        --lexical_mapping_file='${DIR}/data/word2postag.txt,${DIR}/data/word2char.txt' \
        --lexical_embedding_type='postag,char' \
        --embedding_size=256 \
        --lexical_embedding_size='32,64' \
        --support_ingraph=True"
    fi
  done | bash

  if [ ! -f ${OUTPUT_DIR}/out.json ]; then
    python ${DIR}/tools/merge_json_lists.py ${OUTPUT_DIR}/part-?.json > ${OUTPUT_DIR}/out.json
    echo output saved to ${OUTPUT_DIR}/out.json
  fi

  if [ ! -f ${OUTPUT_DIR}/out.eval ]; then
    python ${DIR}/tools/eval/run_evaluations.py --submit ${OUTPUT_DIR}/out.json --ref $VALIDATE_REFERENCE_FILE | tee ${OUTPUT_DIR}/out.eval | grep ^Eval
    echo eval result saved to ${OUTPUT_DIR}/out.eval
  fi
done