README.md

# TensorFlow-Slim image classification library

[TF-slim](https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/slim)
is a new lightweight high-level API of TensorFlow (`tensorflow.contrib.slim`)
for defining, training and evaluating complex
models. This directory contains
code for training and evaluating several widely used Convolutional Neural
Network (CNN) image classification models using TF-slim.
It contains scripts that will allow
you to train models from scratch or fine-tune them from pre-trained network
weights. It also contains code for downloading standard image datasets,
converting them
to TensorFlow's native TFRecord format and reading them in using TF-Slim's
data reading and queueing utilities. You can easily train any model on any of
these datasets, as we demonstrate below. We've also included a
[jupyter notebook](https://github.com/tensorflow/models/blob/master/slim/slim_walkthough.ipynb),
which provides working examples of how to use TF-Slim for image classification.


## Table of contents

<a href="#Install">Installation and setup</a><br>
<a href='#Data'>Preparing the datasets</a><br>
<a href='#Pretrained'>Using pre-trained models</a><br>
<a href='#Training'>Training from scratch</a><br>
<a href='#Tuning'>Fine tuning to a new task</a><br>
<a href='#Eval'>Evaluating performance</a><br>

# Installation
<a id='Install'></a>

In this section, we describe the steps required to install the appropriate
prerequisite packages.

## Installing latest version of TF-slim

As of 8/28/16, the latest [stable release of TF](https://www.tensorflow.org/versions/r0.10/get_started/os_setup.html#pip-installation)
is r0.10, which contains most of TF-Slim but not some later additions. To obtain the
latest version, you must install the most recent nightly build of
TensorFlow. You can find the latest nightly binaries at
[TensorFlow Installation](https://github.com/tensorflow/tensorflow#installation)
in the section that reads "People who are a little more adventurous can
also try our nightly binaries". Copy the link address that corresponds to
the appropriate machine architecture and python version, and pip install
it. For example:

```shell
export TF_BINARY_URL=https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_CONTAINER_TYPE=CPU,TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.10.0rc0-cp27-none-linux_x86_64.whl
sudo pip install --upgrade $TF_BINARY_URL
```

To test this has worked, execute the following command; it should run
without raising any errors.

```
python -c "import tensorflow.contrib.slim as slim; eval = slim.evaluation.evaluate_once"
```

## Installing the TF-slim image models library

To use TF-Slim for image classification, you also have to install
the [TF-Slim image models library](https://github.com/tensorflow/models/tree/master/slim),
which is not part of the core TF library.
To do this, check out the
[tensorflow/models](https://github.com/tensorflow/models/) repository as follows:

```bash
cd $HOME/workspace
git clone https://github.com/tensorflow/models/
```

This will put the TF-Slim image models library in `$HOME/workspace/models/slim`.
(It will also create a directory called
[models/inception](https://github.com/tensorflow/models/tree/master/inception),
which contains an older version of slim; you can safely ignore this.)

To verify that this has worked, execute the following commands; it should run
without raising any errors.

```
cd $HOME/workspace/models/slim
python -c "from nets import cifarnet; mynet = cifarnet.cifarnet"
```


# Preparing the datasets
<a id='Data'></a>

As part of this library, we've included scripts to download several popular
image datasets (listed below) and convert them to slim format.

Dataset | Training Set Size | Testing Set Size | Number of Classes | Comments
:------:|:---------------:|:---------------------:|:-----------:|:-----------:
Flowers|2500 | 2500 | 5 | Various sizes (source: Flickr)
[Cifar10](https://www.cs.toronto.edu/~kriz/cifar.html) | 60k| 10k | 10 |32x32 color
[MNIST](http://yann.lecun.com/exdb/mnist/)| 60k | 10k | 10 | 28x28 gray
[ImageNet](http://www.image-net.org/challenges/LSVRC/2012/)|1.2M| 50k | 1000 | Various sizes

## Downloading and converting to TFRecord format

For each dataset, we'll need to download the raw data and convert it to
TensorFlow's native
[TFRecord](https://www.tensorflow.org/versions/r0.10/api_docs/python/python_io.html#tfrecords-format-details)
format. Each TFRecord contains a
[TF-Example](https://github.com/tensorflow/tensorflow/blob/r0.10/tensorflow/core/example/example.proto)
protocol buffer. Below we demonstrate how to do this for the Flowers dataset.

```shell
$ DATA_DIR=/tmp/data/flowers
$ python download_and_convert_data.py \
    --dataset_name=flowers \
    --dataset_dir="${DATA_DIR}"
```

When the script finishes you will find several TFRecord files created:

```shell
$ ls ${DATA_DIR}
flowers_train-00000-of-00005.tfrecord
...
flowers_train-00004-of-00005.tfrecord
flowers_validation-00000-of-00005.tfrecord
...
flowers_validation-00004-of-00005.tfrecord
labels.txt
```

These represent the training and validation data, sharded over 5 files each.
You will also find the `$DATA_DIR/labels.txt` file which contains the mapping
from integer labels to class names.

You can use the same script to create the mnist and cifar10 datasets.
However, for ImageNet, you have to follow the instructions
[here](https://github.com/tensorflow/models/blob/master/inception/README.md#getting-started).
Note that you first have to sign up for an account at image-net.org.
Also, the download can take several hours, and uses about 500MB.


## Creating a TF-Slim Dataset Descriptor.

Once the TFRecord files have been created, you can easily define a Slim
[Dataset](https://github.com/tensorflow/tensorflow/blob/r0.10/tensorflow/contrib/slim/python/slim/data/dataset.py),
which stores pointers to the data file, as well as various other pieces of
metadata, such as the class labels, the train/test split, and how to parse the
TFExample protos. We have included the TF-Slim Dataset descriptors
for
[Cifar10](https://github.com/tensorflow/models/blob/master/slim/datasets/cifar10.py),
[ImageNet](https://github.com/tensorflow/models/blob/master/slim/datasets/imagenet.py),
[Flowers](https://github.com/tensorflow/models/blob/master/slim/datasets/flowers.py),
and
[MNIST](https://github.com/tensorflow/models/blob/master/slim/datasets/mnist.py).
An example of how to load data using a TF-Slim dataset descriptor using a
TF-Slim
[DatasetDataProvider](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/data/dataset_data_provider.py)
is found below:

```python
import tensorflow as tf
from datasets import flowers

slim = tf.contrib.slim

# Selects the 'validation' dataset.
dataset = flowers.get_split('validation', DATA_DIR)

# Creates a TF-Slim DataProvider which reads the dataset in the background
# during both training and testing.
provider = slim.dataset_data_provider.DatasetDataProvider(dataset)
[image, label] = provider.get(['image', 'label'])
```


# Pre-trained Models
<a id='Pretrained'></a>

Neural nets work best when they have many parameters, making them powerful
function approximators.
However, this  means they must be trained on very large datasets. Because
training models from scratch can be a very computationally intensive process
requiring days or even weeks, we provide various pre-trained models,
as listed below. These CNNs have been trained on the
[ILSVRC-2012-CLS](http://www.image-net.org/challenges/LSVRC/2012/)
image classification dataset.

In the table below, we list each model, the corresponding
TensorFlow model file, the link to the model checkpoint, and the top 1 and top 5
accuracy (on the imagenet test set).
Note that the VGG and ResNet parameters have been converted from their original
caffe formats
([here](https://github.com/BVLC/caffe/wiki/Model-Zoo#models-used-by-the-vgg-team-in-ilsvrc-2014)
and
[here](https://github.com/KaimingHe/deep-residual-networks)),
whereas the Inception parameters have been trained internally at
Google. Also be aware that these accuracies were computed by evaluating using a
single image crop. Some academic papers report higher accuracy by using multiple
crops at multiple scales.

Model | TF-Slim File | Checkpoint | Top-1 Accuracy| Top-5 Accuracy |
:----:|:------------:|:----------:|:-------:|:--------:|
[Inception V1](http://arxiv.org/abs/1409.4842v1)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/inception_v1.py)|[inception_v1.tar.gz](http://download.tensorflow.org/models/inception_v1_2016_08_28.tar.gz)|69.8|89.6|
[Inception V2](http://arxiv.org/abs/1502.03167)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/inception_v2.py)|[inception_v2.tar.gz](http://download.tensorflow.org/models/inception_v2_2016_08_28.tar.gz)|73.9|91.8|
[Inception V3](http://arxiv.org/abs/1512.00567)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/inception_v3.py)|[inception_v3.tar.gz](http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz)|78.0|93.9|
[ResNet 50](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/resnet_v1.py)|[resnet_v1_50.tar.gz](http://download.tensorflow.org/models/resnet_v1_50_2016_08_28.tar.gz)|75.2|92.2|
[ResNet 101](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/resnet_v1.py)|[resnet_v1_101.tar.gz](http://download.tensorflow.org/models/resnet_v1_101_2016_08_28.tar.gz)|76.4|92.9|
[ResNet 152](https://arxiv.org/abs/1512.03385)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/resnet_v1.py)|[resnet_v1_152.tar.gz](http://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz)|76.8|93.2|
[VGG 16](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/vgg.py)|[vgg_16.tar.gz](http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz)|71.5|89.8|
[VGG 19](http://arxiv.org/abs/1409.1556.pdf)|[Code](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/slim/python/slim/nets/vgg.py)|[vgg_19.tar.gz](http://download.tensorflow.org/models/vgg_19_2016_08_28.tar.gz)|71.1|89.8|


Here is an example of how to download the Inception V3 checkpoint:

```shell
$ CHECKPOINT_DIR=/tmp/checkpoints
$ mkdir ${CHECKPOINT_DIR}
$ wget http://download.tensorflow.org/models/inception_v3_2016_08_28.tar.gz
$ tar -xvf inception_v3_2016_08_28.tar.gz
$ mv inception_v3.ckpt ${CHECKPOINT_DIR}
$ rm inception_v3_2016_08_28.tar.gz
```



# Training a model from scratch.
<a id='Training'></a>

We provide an easy way to train a model from scratch using any TF-Slim dataset.
The following example demonstrates how to train Inception V3 using the default
parameters on the ImageNet dataset.

```shell
DATASET_DIR=/tmp/imagenet
TRAIN_DIR=/tmp/train_logs
python train_image_classifier.py \
    --train_dir=${TRAIN_DIR} \
    --dataset_name=imagenet \
    --dataset_split_name=train \
    --dataset_dir=${DATASET_DIR} \
    --model_name=inception_v3
```

This process may take several days, depending on your hardware setup.
For convenience, we provide a way to train a model on multiple GPUs,
and/or multiple CPUs, either synchrononously or asynchronously.
See [model_deploy](https://github.com/tensorflow/models/blob/master/slim/deployment/model_deploy.py)
for details.


# Fine-tuning a model from an existing checkpoint
<a id='Tuning'></a>

Rather than training from scratch, we'll often want to start from a pre-trained
model and fine-tune it.
To indicate a checkpoint from which to fine-tune, we'll call training with
the `--checkpoint_path` flag and assign it an absolute path to a checkpoint
file.

When fine-tuning a model, we need to be careful about restoring checkpoint
weights. In particular, when we fine-tune a model on a new task with a different
number of output labels, we wont be able restore the final logits (classifier)
layer. For this, we'll use the `--checkpoint_exclude_scopes` flag. This flag
hinders certain variables from being loaded. When fine-tuning on a
classification task using a different number of classes than the trained model,
the new model will have a final 'logits' layer whose dimensions differ from the
pre-trained model. For example, if fine-tuning an ImageNet-trained model on
Flowers, the pre-trained logits layer will have dimensions `[2048 x 1001]` but
our new logits layer will have dimensions `[2048 x 5]`. Consequently, this
flag indicates to TF-Slim to avoid loading these weights from the checkpoint.

Keep in mind that warm-starting from a checkpoint affects the model's weights
only during the initialization of the model. Once a model has started training,
a new checkpoint will be created in `${TRAIN_DIR}`. If the fine-tuning
training is stopped and restarted, this new checkpoint will be the one from
which weights are restored and not the `${checkpoint_path}$`. Consequently,
the flags `--checkpoint_path` and `--checkpoint_exclude_scopes` are only used
during the `0-`th global step (model initialization). Typically for fine-tuning
one only want train a sub-set of layers, so the flag `--trainable_scopes` allows
to specify which subsets of layers should trained, the rest would remain frozen.

Below we give an example of
[fine-tuning inception-v3 on flowers](https://github.com/tensorflow/models/blob/master/slim/scripts/finetune_inception_v3_on_flowers.sh),
inception_v3  was trained on ImageNet with 1000 class labels, but the flowers
dataset only have 5 classes. Since the dataset is quite small we will only train
the new layers.


```shell
$ DATASET_DIR=/tmp/flowers
$ TRAIN_DIR=/tmp/flowers-models/inception_v3
$ CHECKPOINT_PATH=/tmp/my_checkpoints/inception_v3.ckpt
$ python train_image_classifier.py \
    --train_dir=${TRAIN_DIR} \
    --dataset_dir=${DATASET_DIR} \
    --dataset_name=flowers \
    --dataset_split_name=train \
    --model_name=inception_v3 \
    --checkpoint_path=${CHECKPOINT_PATH} \
    --checkpoint_exclude_scopes=InceptionV3/Logits,InceptionV3/AuxLogits/Logits \
    --trainable_scopes=InceptionV3/Logits,InceptionV3/AuxLogits/Logits
```



# Evaluating performance of a model
<a id='Eval'></a>

To evaluate the performance of a model (whether pretrained or your own),
you can use the eval_image_classifier.py script, as shown below.

Below we give an example of downloading the pretrained inception model and
evaluating it on the imagenet dataset.

```shell
CHECKPOINT_FILE = ${CHECKPOINT_DIR}/inception_v3.ckpt  # Example
$ python eval_image_classifier.py \
    --alsologtostderr \
    --checkpoint_path=${CHECKPOINT_FILE} \
    --dataset_dir=${DATASET_DIR} \
    --dataset_name=imagenet \
    --dataset_split_name=validation \
    --model_name=inception_v3
```



# Troubleshooting

#### The model runs out of CPU memory.

See
[Model Runs out of CPU memory](https://github.com/tensorflow/models/tree/master/inception#the-model-runs-out-of-cpu-memory).

#### The model runs out of GPU memory.

See
[Adjusting Memory Demands](https://github.com/tensorflow/models/tree/master/inception#adjusting-memory-demands).

#### The model training results in NaN's.

See
[Model Resulting in NaNs](https://github.com/tensorflow/models/tree/master/inception#the-model-training-results-in-nans).

#### The ResNet and VGG Models have 1000 classes but the ImageNet dataset has 1001

The ImageNet dataset provied has an empty background class which was can be used
to fine-tune the model to other tasks. If you try training or fine-tuning the
VGG or ResNet models using the ImageNet dataset, you might encounter the
following error:

```bash
InvalidArgumentError: Assign requires shapes of both tensors to match. lhs shape= [1001] rhs shape= [1000]
```
This is due to the fact that the VGG and ResNet final layers have only 1000
outputs rather than 1001.

To fix this issue, you can set the `--labels_offsets=1` flag. This results in
the ImageNet labels being shifted down by one:


#### I wish to train a model with a different image size.

The preprocessing functions all take `height` and `width` as parameters. You
can change the default values using the following snippet:

```python
image_preprocessing_fn = preprocessing_factory.get_preprocessing(
    preprocessing_name,
    height=MY_NEW_HEIGHT,
    width=MY_NEW_WIDTH,
    is_training=True)
```

#### What hardware specification are these hyper-parameters targeted for?

See
[Hardware Specifications](https://github.com/tensorflow/models/tree/master/inception#what-hardware-specification-are-these-hyper-parameters-targeted-for).