FC2カウンター FPGAの部屋 Edge TPU の”Retrain an image classification model”をやってみる1(画像分類モデルを転移学習)
FC2ブログ

FPGAやCPLDの話題やFPGA用のツールの話題などです。 マニアックです。 日記も書きます。

FPGAの部屋

FPGAの部屋の有用と思われるコンテンツのまとめサイトを作りました。Xilinx ISEの初心者の方には、FPGAリテラシーおよびチュートリアルのページをお勧めいたします。

Edge TPU の”Retrain an image classification model”をやってみる1(画像分類モデルを転移学習)

Corel Beta の”TensorFlow models on the Edge TPU”のリンクに”Retrain an image classification model”があったので、やってみようと思う。

このチュートリアルでは、量子化された MobileNet V1モデルを再学習してさまざまな種類の花を認識する方法をやってみるそうだ。
具体的には転移学習で、最後の層だけの再学習と全部の層の再学習の2つを試す事ができるそうだ。また、Docker を使ってチュートリアルを実行する手順になっているのも魅力的だ。まずは、最後の層だけ再学習をやってみよう。

まずは、CLASSIFY_DIR を定義しよう。CLASSIFY_DIR は ~/python-tflite-source/edgetpu/classify に設定する。
export CLASSIFY_DIR=/home/masaaki/python-tflite-source/edgetpu/classify

~/python-tflite-source/edgetpu/ ディレクトリの下に classify ディレクトリを生成する。
mkdir classify
cd classify


Docker ファイルをダウンロードする。
wget -O Dockerfile "http://storage.googleapis.com/cloud-iot-edge-pretrained-models/docker/classify_docker"
Edge_TPU_24_190312.png

docker build を行った。
docker build - < Dockerfile --tag classify-tutorial
Edge_TPU_25_190312.png

Docker コンテナを起動する。
docker run --name edgetpu-classify \
--rm -it --privileged -p 6006:6006 \
--mount type=bind,src=${CLASSIFY_DIR},dst=/tensorflow/models/research/slim/transfer_learn \
classify-tutorial

実行すると

root@abedb1f7136c:/tensorflow/models/research/slim#

のプロンプトが表示された。
Edge_TPU_27_190313.png

flowersデータセットをダウンロードしてTFRecord形式に変換する必要があるそうだ。
./prepare_checkpoint_and_dataset.sh --network_type mobilenet_v1
Edge_TPU_28_190313.png

ログを示す。

root@abedb1f7136c:/tensorflow/models/research/slim# ./prepare_checkpoint_and_dataset.sh --network_type mobilenet_v1
+ network_type=mobilenet_v1
+ [[ 2 -gt 0 ]]
+ case "$1" in
+ network_type=mobilenet_v1
+ shift 2
+ [[ 0 -gt 0 ]]
+ source /tensorflow/models/research/slim/constants.sh
++ declare -A ckpt_link_map
++ declare -A ckpt_name_map
++ declare -A image_size_map
++ declare -A scopes_map
++ declare -A input_tensors_map
++ declare -A output_tensors_map
++ ckpt_link_map["mobilenet_v1"]=http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz
++ ckpt_link_map["mobilenet_v2"]=http://download.tensorflow.org/models/tflite_11_05_08/mobilenet_v2_1.0_224_quant.tgz
++ ckpt_link_map["inception_v1"]=http://download.tensorflow.org/models/inception_v1_224_quant_20181026.tgz
++ ckpt_link_map["inception_v2"]=http://download.tensorflow.org/models/inception_v2_224_quant_20181026.tgz
++ ckpt_link_map["inception_v3"]=http://download.tensorflow.org/models/tflite_11_05_08/inception_v3_quant.tgz
++ ckpt_link_map["inception_v4"]=http://download.tensorflow.org/models/inception_v4_299_quant_20181026.tgz
++ ckpt_name_map["mobilenet_v1"]=mobilenet_v1_1.0_224_quant
++ ckpt_name_map["mobilenet_v2"]=mobilenet_v2_1.0_224_quant
++ ckpt_name_map["inception_v1"]=inception_v1_224_quant
++ ckpt_name_map["inception_v2"]=inception_v2_224_quant
++ ckpt_name_map["inception_v3"]=inception_v3_quant
++ ckpt_name_map["inception_v4"]=inception_v4_299_quant
++ image_size_map["mobilenet_v1"]=224
++ image_size_map["mobilenet_v2"]=224
++ image_size_map["inception_v1"]=224
++ image_size_map["inception_v2"]=224
++ image_size_map["inception_v3"]=299
++ image_size_map["inception_v4"]=299
++ scopes_map["mobilenet_v1"]=MobilenetV1/Logits
++ scopes_map["mobilenet_v2"]=MobilenetV2/Logits
++ scopes_map["inception_v1"]=InceptionV1/Logits
++ scopes_map["inception_v2"]=InceptionV2/Logits
++ scopes_map["inception_v3"]=InceptionV3/Logits,InceptionV3/AuxLogits
++ scopes_map["inception_v4"]=InceptionV4/Logits,InceptionV4/AuxLogits
++ input_tensors_map["mobilenet_v1"]=input
++ input_tensors_map["mobilenet_v2"]=input
++ input_tensors_map["inception_v1"]=input
++ input_tensors_map["inception_v2"]=input
++ input_tensors_map["inception_v3"]=input
++ input_tensors_map["inception_v4"]=input
++ output_tensors_map["mobilenet_v1"]=MobilenetV1/Predictions/Reshape_1
++ output_tensors_map["mobilenet_v2"]=MobilenetV2/Predictions/Softmax
++ output_tensors_map["inception_v1"]=InceptionV1/Logits/Predictions/Softmax
++ output_tensors_map["inception_v2"]=InceptionV2/Predictions/Reshape_1
++ output_tensors_map["inception_v3"]=InceptionV3/Predictions/Reshape_1
++ output_tensors_map["inception_v4"]=InceptionV4/Logits/Predictions
++ SLIM_DIR=/tensorflow/models/research/slim
++ LEARN_DIR=/tensorflow/models/research/slim/transfer_learn
++ CKPT_DIR=/tensorflow/models/research/slim/transfer_learn/ckpt
++ DATASET_DIR=/tensorflow/models/research/slim/transfer_learn/flowers
++ TRAIN_DIR=/tensorflow/models/research/slim/transfer_learn/train
++ OUTPUT_DIR=/tensorflow/models/research/slim/transfer_learn/models
+ echo 'PREPARING checkpoint ...'
PREPARING checkpoint ...
+ mkdir -p /tensorflow/models/research/slim/transfer_learn
+ mkdir /tensorflow/models/research/slim/transfer_learn/ckpt
+ cd /tensorflow/models/research/slim/transfer_learn/ckpt
+ ckpt_link=http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz
+ ckpt_name=mobilenet_v1_1.0_224_quant
+ wget -O mobilenet_v1_1.0_224_quant.tgz http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz
--2019-03-12 11:22:56--  http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz
Resolving download.tensorflow.org (download.tensorflow.org)... 172.217.25.208, 2404:6800:4004:818::2010
Connecting to download.tensorflow.org (download.tensorflow.org)|172.217.25.208|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 35069912 (33M) [application/x-compressed-tar]
Saving to: 'mobilenet_v1_1.0_224_quant.tgz'

mobilenet_v1_1.0_22 100%[===================>]  33.44M  9.83MB/s    in 3.4s    

2019-03-12 11:23:00 (9.83 MB/s) - 'mobilenet_v1_1.0_224_quant.tgz' saved [35069912/35069912]

+ tar zxvf mobilenet_v1_1.0_224_quant.tgz
./
./mobilenet_v1_1.0_224_quant.ckpt.index
./mobilenet_v1_1.0_224_quant_eval.pbtxt
./mobilenet_v1_1.0_224_quant_info.txt
./mobilenet_v1_1.0_224_quant.ckpt.data-00000-of-00001
./mobilenet_v1_1.0_224_quant.tflite
./mobilenet_v1_1.0_224_quant.ckpt.meta
./mobilenet_v1_1.0_224_quant_frozen.pb
+ echo 'PREPARING dataset ...'
PREPARING dataset ...
+ mkdir /tensorflow/models/research/slim/transfer_learn/flowers
+ cd /tensorflow/models/research/slim
+ python download_and_convert_data.py --dataset_name=flowers --dataset_dir=/tensorflow/models/research/slim/transfer_learn/flowers
>> Downloading flower_photos.tgz 100.0%
Successfully downloaded flower_photos.tgz 228813984 bytes.
2019-03-12 11:23:25.559151: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
>> Converting image 3320/3320 shard 4
>> Converting image 350/350 shard 4

Finished converting the Flowers dataset!
+ echo 'CHECKPOINT and dataset available in /tensorflow/models/research/slim/transfer_learn'
CHECKPOINT and dataset available in /tensorflow/models/research/slim/transfer_learn



画像分類モデルを再学習する。最後の層だけを再学習してみよう。
./start_training.sh --network_type mobilenet_v1
Edge_TPU_26_190312.png

ログを示す。

root@abedb1f7136c:/tensorflow/models/research/slim# ./start_training.sh --network_type mobilenet_v1
+ network_type=mobilenet_v1
+ train_whole_model=false
+ train_steps=300
+ quantize_delay=100
+ [[ 2 -gt 0 ]]
+ case "$1" in
+ network_type=mobilenet_v1
+ shift 2
+ [[ 0 -gt 0 ]]
+ source /tensorflow/models/research/slim/constants.sh
++ declare -A ckpt_link_map
++ declare -A ckpt_name_map
++ declare -A image_size_map
++ declare -A scopes_map
++ declare -A input_tensors_map
++ declare -A output_tensors_map
++ ckpt_link_map["mobilenet_v1"]=http://download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224_quant.tgz
++ ckpt_link_map["mobilenet_v2"]=http://download.tensorflow.org/models/tflite_11_05_08/mobilenet_v2_1.0_224_quant.tgz
++ ckpt_link_map["inception_v1"]=http://download.tensorflow.org/models/inception_v1_224_quant_20181026.tgz
++ ckpt_link_map["inception_v2"]=http://download.tensorflow.org/models/inception_v2_224_quant_20181026.tgz
++ ckpt_link_map["inception_v3"]=http://download.tensorflow.org/models/tflite_11_05_08/inception_v3_quant.tgz
++ ckpt_link_map["inception_v4"]=http://download.tensorflow.org/models/inception_v4_299_quant_20181026.tgz
++ ckpt_name_map["mobilenet_v1"]=mobilenet_v1_1.0_224_quant
++ ckpt_name_map["mobilenet_v2"]=mobilenet_v2_1.0_224_quant
++ ckpt_name_map["inception_v1"]=inception_v1_224_quant
++ ckpt_name_map["inception_v2"]=inception_v2_224_quant
++ ckpt_name_map["inception_v3"]=inception_v3_quant
++ ckpt_name_map["inception_v4"]=inception_v4_299_quant
++ image_size_map["mobilenet_v1"]=224
++ image_size_map["mobilenet_v2"]=224
++ image_size_map["inception_v1"]=224
++ image_size_map["inception_v2"]=224
++ image_size_map["inception_v3"]=299
++ image_size_map["inception_v4"]=299
++ scopes_map["mobilenet_v1"]=MobilenetV1/Logits
++ scopes_map["mobilenet_v2"]=MobilenetV2/Logits
++ scopes_map["inception_v1"]=InceptionV1/Logits
++ scopes_map["inception_v2"]=InceptionV2/Logits
++ scopes_map["inception_v3"]=InceptionV3/Logits,InceptionV3/AuxLogits
++ scopes_map["inception_v4"]=InceptionV4/Logits,InceptionV4/AuxLogits
++ input_tensors_map["mobilenet_v1"]=input
++ input_tensors_map["mobilenet_v2"]=input
++ input_tensors_map["inception_v1"]=input
++ input_tensors_map["inception_v2"]=input
++ input_tensors_map["inception_v3"]=input
++ input_tensors_map["inception_v4"]=input
++ output_tensors_map["mobilenet_v1"]=MobilenetV1/Predictions/Reshape_1
++ output_tensors_map["mobilenet_v2"]=MobilenetV2/Predictions/Softmax
++ output_tensors_map["inception_v1"]=InceptionV1/Logits/Predictions/Softmax
++ output_tensors_map["inception_v2"]=InceptionV2/Predictions/Reshape_1
++ output_tensors_map["inception_v3"]=InceptionV3/Predictions/Reshape_1
++ output_tensors_map["inception_v4"]=InceptionV4/Logits/Predictions
++ SLIM_DIR=/tensorflow/models/research/slim
++ LEARN_DIR=/tensorflow/models/research/slim/transfer_learn
++ CKPT_DIR=/tensorflow/models/research/slim/transfer_learn/ckpt
++ DATASET_DIR=/tensorflow/models/research/slim/transfer_learn/flowers
++ TRAIN_DIR=/tensorflow/models/research/slim/transfer_learn/train
++ OUTPUT_DIR=/tensorflow/models/research/slim/transfer_learn/models
+ mkdir /tensorflow/models/research/slim/transfer_learn/train
+ image_size=224
+ ckpt_name=mobilenet_v1_1.0_224_quant
+ scopes=MobilenetV1/Logits
+ [[ false == \t\r\u\e ]]
+ echo 'TRAINING last few layers ...'
TRAINING last few layers ...
+ python train_image_classifier.py --train_dir=/tensorflow/models/research/slim/transfer_learn/train --dataset_name=flowers --dataset_split_name=train --dataset_dir=/tensorflow/models/research/slim/transfer_learn/flowers --model_name=mobilenet_v1 --checkpoint_path=/tensorflow/models/research/slim/transfer_learn/ckpt/mobilenet_v1_1.0_224_quant.ckpt --max_number_of_steps=300 --batch_size=100 --learning_rate=0.01 --learning_rate_decay_type=fixed --save_interval_secs=60 --save_summaries_secs=60 --log_every_n_steps=20 --optimizer=sgd --weight_decay=0.00004 --quantize_delay=100 --clone_on_cpu --train_image_size=224 --checkpoint_exclude_scopes=MobilenetV1/Logits --trainable_scopes=MobilenetV1/Logits
WARNING:tensorflow:From train_image_classifier.py:413: create_global_step (from tensorflow.contrib.framework.python.ops.variables) is deprecated and will be removed in a future version.
Instructions for updating:
Please switch to tf.train.create_global_step
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/python/training/input.py:187: __init__ (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/python/training/input.py:187: add_queue_runner (from tensorflow.python.training.queue_runner_impl) is deprecated and will be removed in a future version.
Instructions for updating:
To construct input pipelines, use the `tf.data` module.
WARNING:tensorflow:From train_image_classifier.py:481: softmax_cross_entropy (from tensorflow.contrib.losses.python.losses.loss_ops) is deprecated and will be removed after 2016-12-30.
Instructions for updating:
Use tf.losses.softmax_cross_entropy instead. Note that the order of the logits and labels arguments has been changed.
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/contrib/losses/python/losses/loss_ops.py:398: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.
Instructions for updating:

Future major versions of TensorFlow will allow gradients to flow
into the labels input on backprop by default.

See `tf.nn.softmax_cross_entropy_with_logits_v2`.

WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/contrib/losses/python/losses/loss_ops.py:399: compute_weighted_loss (from tensorflow.contrib.losses.python.losses.loss_ops) is deprecated and will be removed after 2016-12-30.
Instructions for updating:
Use tf.losses.compute_weighted_loss instead.
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/contrib/losses/python/losses/loss_ops.py:147: add_loss (from tensorflow.contrib.losses.python.losses.loss_ops) is deprecated and will be removed after 2016-12-30.
Instructions for updating:
Use tf.losses.add_loss instead.
INFO:tensorflow:Fine-tuning from /tensorflow/models/research/slim/transfer_learn/ckpt/mobilenet_v1_1.0_224_quant.ckpt
WARNING:tensorflow:From /usr/local/lib/python2.7/dist-packages/tensorflow/contrib/slim/python/slim/learning.py:737: __init__ (from tensorflow.python.training.supervisor) is deprecated and will be removed in a future version.
Instructions for updating:
Please switch to tf.train.MonitoredTrainingSession
2019-03-12 11:27:30.706494: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
INFO:tensorflow:Restoring parameters from /tensorflow/models/research/slim/transfer_learn/ckpt/mobilenet_v1_1.0_224_quant.ckpt
INFO:tensorflow:Running local_init_op.
INFO:tensorflow:Done running local_init_op.
INFO:tensorflow:Starting Session.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Starting Queues.
INFO:tensorflow:global_step/sec: 0
INFO:tensorflow:Recording summary at step 1.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.134039
INFO:tensorflow:Recording summary at step 9.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.149296
INFO:tensorflow:Recording summary at step 18.
INFO:tensorflow:global step 20: loss = 1.8288 (5.751 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.150273
INFO:tensorflow:Recording summary at step 27.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.150492
INFO:tensorflow:Recording summary at step 36.
INFO:tensorflow:global step 40: loss = 1.2239 (5.355 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.147502
INFO:tensorflow:Recording summary at step 45.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.15272
INFO:tensorflow:Recording summary at step 54.
INFO:tensorflow:global step 60: loss = 1.0667 (5.793 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.149623
INFO:tensorflow:Recording summary at step 63.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.13363
INFO:tensorflow:Recording summary at step 72.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global_step/sec: 0.14869
INFO:tensorflow:global step 80: loss = 0.8747 (9.908 sec/step)
INFO:tensorflow:Recording summary at step 80.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 89.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 99.
INFO:tensorflow:global step 100: loss = 0.9789 (5.818 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 108.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 117.
INFO:tensorflow:global step 120: loss = 0.7836 (5.771 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 126.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 135.
INFO:tensorflow:global step 140: loss = 0.7582 (5.641 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 144.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 153.
INFO:tensorflow:global step 160: loss = 0.7289 (5.561 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 162.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 171.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:global step 180: loss = 0.7996 (8.771 sec/step)
INFO:tensorflow:Recording summary at step 180.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 190.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 199.
INFO:tensorflow:global step 200: loss = 0.6564 (6.531 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 208.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 217.
INFO:tensorflow:global step 220: loss = 0.7724 (5.751 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 226.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 236.
INFO:tensorflow:global step 240: loss = 0.7227 (5.387 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 245.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 255.
INFO:tensorflow:global step 260: loss = 0.7362 (5.324 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 264.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 274.
INFO:tensorflow:global step 280: loss = 0.5858 (5.351 sec/step)
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 283.
INFO:tensorflow:Saving checkpoint to path /tensorflow/models/research/slim/transfer_learn/train/model.ckpt
INFO:tensorflow:Recording summary at step 293.
INFO:tensorflow:global step 300: loss = 0.6365 (5.394 sec/step)
INFO:tensorflow:Stopping Training.
INFO:tensorflow:Finished training! Saving model to disk.

  1. 2019年03月13日 06:37 |
  2. Edge TPU
  3. | トラックバック:0
  4. | コメント:0

コメント

コメントの投稿


管理者にだけ表示を許可する

トラックバック URL
http://marsee101.blog.fc2.com/tb.php/4471-31e76fdd
この記事にトラックバックする(FC2ブログユーザー)