Github庫：https://github.com/tensorflow/models/tree/master/research/im2txt

場景：給一張圖片，描述圖片內(nèi)容。如下圖：

環(huán)境：ubuntu，記住用戶必須對目錄有權(quán)限。

1、下載開源包：git clone?https://github.com/tensorflow/models.git? ?

? ? ? 進(jìn)入Im2txt子項目：cd?research/im2txt

2、安裝依賴包：

First ensure that you have installed the following required packages:

• Bazel?(instructions)
• Python 2.7
• TensorFlow?1.0 or greater (instructions)
• NumPy?(instructions)
• Natural Language Toolkit (NLTK):
  • First install NLTK (instructions)
  • Then install the NLTK data package "punkt" (instructions)
• Unzip

1）Bazel：?我選擇Installing using binary installer方式安裝。

? ? ?參考網(wǎng)址：https://docs.bazel.build/versions/master/install-ubuntu.html

2）NLTK：sudo?pip2 install?-U?nltk

? ? ?數(shù)據(jù)包安裝：進(jìn)入python2

>>>import nltk >>> nltk.download("punkt")

3、準(zhǔn)備訓(xùn)練數(shù)據(jù)

3.1??MSCOCO數(shù)據(jù)集

bazel build //im2txt:download_and_preprocess_mscoco bazel-bin/im2txt/download_and_preprocess_mscoco "im2txt/data/mscoco"

準(zhǔn)備成功如下圖：

3.2 已訓(xùn)練CNN模型Inception v3 Checkpoint

wget "http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz" tar -xvf "inception_v3_2016_08_28.tar.gz" -C im2txt/data rm "inception_v3_2016_08_28.tar.gz"

4、訓(xùn)練模型

bazel build -c opt //im2txt/...

nohup bazel-bin/im2txt/train \
? --input_file_pattern="im2txt/data/mscoco/train-?????-of-00256" \
? --inception_checkpoint_file="im2txt/data/inception_v3.ckpt" \
? --train_dir="im2txt/model/train" \
? --train_inception=false \
? --number_of_steps=1000000 &

evaluation script和Fine Tune the Inception v3 Model我這里沒有去運(yùn)行。有需要可參照官方介紹進(jìn)行。訓(xùn)練完成如下圖。

5、圖片生成文本

bazel build -c opt //im2txt:run_inference bazel-bin/im2txt/run_inference \--checkpoint_path=im2txt/model/train \--vocab_file=im2txt/data/mscoco/word_counts.txt \--input_files=im2txt/data/mscoco/raw-data/val2014/COCO_val2014_000000051008.jpg

?結(jié)果：

Captions for image COCO_val2014_000000051008.jpg:0) a cat laying on top of a laptop computer . (p=0.010717)1) a cat laying on top of a laptop keyboard . (p=0.001352)2) a cat laying on top of a laptop on a bed . (p=0.000764)

總結(jié)

以上是生活随笔為你收集整理的Google开源库Image Captioning部署记录的全部內(nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯，歡迎將生活随笔推薦給好友。