Tensorflow inception-v4 Classify Image - tensorflow

Question

I use TF-slim inception-v4 training a model from scratch.
python train_image_classifier.py \
--train_dir=${TRAIN_DIR} \
--dataset_name=mydata \
--dataset_split_name=train \
--dataset_dir=${DATASET_DIR} \
--model_name=inception_v4 \
--clone_on_cpu=true \
--max_number_of_steps=1000 \
--log_every_n_steps=100
# Run evaluation.
python eval_image_classifier.py \
--checkpoint_path=${TRAIN_DIR} \
--eval_dir=${TRAIN_DIR} \
--dataset_name=mydata \
--dataset_split_name=validation \
--dataset_dir=${DATASET_DIR} \
--model_name=inception_v4 \
--batch_size=32
# # # Fine-tune all the new layers for 500 steps.
python train_image_classifier.py \
--train_dir=${TRAIN_DIR}/all \
--dataset_name=mydata \
--dataset_split_name=train \
--dataset_dir=${DATASET_DIR} \
--model_name=inception_v4 \
--clone_on_cpu=true \
--checkpoint_path=${TRAIN_DIR} \
--max_number_of_steps=1000 \
--log_every_n_steps=100 \
--batch_size=32 \
--learning_rate=0.0001 \
--learning_rate_decay_type=fixed \
--save_interval_secs=600 \
--save_summaries_secs=600 \
--optimizer=rmsprop \
--weight_decay=0.00004
then freeze the graph:
python export_inference_graph.py \
--alsologtostderr \
--model_name=inception_v4 \
--is_training=True \
--labels_offset=999 \
--output_file=${OUTPUT_DIR}/unfrozen_inception_v4_graph.pb \
--dataset_dir=${DATASET_DIR}
#NEWEST_CHECKPOINT=$(cat ${TRAIN_DIR}/all/checkpoint |head -n1|awk -F\" '{print $2}')
NEWEST_CHECKPOINT=$(ls -t1 ${TRAIN_DIR}/all|grep model.ckpt |head -n1)
echo ${NEWEST_CHECKPOINT%.*}
python ${OUTPUT_DIR}/tensorflow/tensorflow/python/tools/freeze_graph.py \
--input_graph=${OUTPUT_DIR}/unfrozen_inception_v4_graph.pb \
--input_checkpoint=${TRAIN_DIR}/all/${NEWEST_CHECKPOINT%.*} \
--input_binary=true \
--output_graph=${OUTPUT_DIR}/frozen_inception_v4.pb \
--output_node_names=InceptionV4/Logits/Predictions \
--input_meta_graph=True
After all this, I got a frozen_inception_v4.pb file.
for this example https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/label_image/label_image.py
what is the input layer for inception_v4 ?
Does anyone know how to solve this?

Answer 1

That depends on the particular implementation of slim you used. Look where they define the input and see what is the name of that tensor.

Answer 2

Try this:
bazel build tensorflow/tools/graph_transforms:summarize_graph
bazel-bin/tensorflow/tools/graph_transforms/summarize_graph \
--in_graph=/path/to/your_frozen.pb
It will show possible input and output layer