Training OverlapSegmentationNet with GPU acceleration on google-colaboratory

 OverlapSegmentationNet is a python implementation of Unet intented to resolve pairs of overlapping chromosomes by semantic segmentation (See previous post).


OverlapSegmentationNet can now be trained on the google colaboratory platform with GPU acceleration.

The whole files and data are available from google drive. Wouldn't have been able to set-up without the following article.

Paper after NIPS 2017 Machine Learning for Health

The article is available on arxiv

Bad predictions from OverlapSegmentationNet after 14 epochs - Solution by Saleh, Saad & Isa

The problems mentioned bellow seems to have solved by Saleh, Saad & Isa in Procedia Computer Science 159 (2019) 524–533

OverlapSegmentationNet is a UNet model implemented with Keras. The network was trained for 14 epochs (~8 hours on a GT740M GPU), this is not enough to make a good prediction but it is enough to play with the model to understand how a prediction from a single image looks like.

Dataset

The dataset used for training was the first proposed on kaggle. It consists in 13434 pairs of greyscale/ground truth images of size 88x88. Each greyscaled image has a pair of overlapping chromosomes. The overlaps were synthetically generated from images of single chromosomes:

Making a prediction from a single grey scaled image:

The aim is to predict the segmentation labels with a Keras model (OverlapSegmentationNet). The 88x88 greyscaled image must be converted into a 4D numpy array (1, 88, 88, 1).

A prediction is then 4D array of shape ( 1,88,88,4), since there's one component for each label (background, chromosome 1, chromosome2, overlap). A prediction looks like:

where each component is a floating-point image. With another chromosome, we have:

A prediction takes ~ 40 ms for one low resolution image.

Prediction made on an image which do not belong to the train dataset:

A prediction was made on an image belonging to a validation dataset:

To be used for prediction, the image was down sampled and recut to fit 88x88 (and its corresponding ground truth segmentation too):

Clearly, with so few epochs the prediction is bad:

The one-hot encoded prediction can be thresholded and back converted into a multilabel segmentation and compared to the ground truth segmentation:

Notebook is here:

More details are available in the notebook:

 

Training OverlapSegmentationNet with gpu acceleration on a Ubuntu powered laptop.

Set-up

Harnessing the gpu 1 : nvidia driver

It is  supposed that a fresh ubuntu 16.04 is set-up.

Install Nvidia proprietary driver from the system parametres menu:

The NVidia installed files, with synaptic package manager, are:

Nvidia acceleration can be switched on graphically after having installed nvidia-settings :

 Check driver installation from a terminal by starting glxgears:

When NVIDIA mode is selected, glxgears performances should increase.

Harnessing the gpu 2: install CUDA 8, cudnn 5.1

CUDA8 was installed from a .deb file available from nvidia's site:

Check cuda installation using the "./deviceQuerry" method (see the end of the post).

The Cudnn 5.1 library was installed by copying the files at the right places (Installing cudnn 6 from .deb archive wasn't a good idea. )

 

Install Tensorflow and Keras 

There's different ways to install tensorflow. Here, both tensorflow with gpu support and keras were installed  in a virtual environnement with pip.

Tensorflow 1.2.1 and Keras 2.0.6 were installed .

h5py was installed from pip too.

Don't forget at the end from a terminal to :

$source .bashrc

Installation of OverlapSegmentationNet:

OverlapSegmentationNet is an implementation of UNet by Dr Hu .

Download image-segmentation-chromosomes or clone it from repository with git.

The image-segmentation-chromosomes directory contains:

$ ls
code  images    README.md

Dataset:

The first version of the low resolution dataset was used. Download it and move it into the code directory:

$ ls

create_histogram_visualisation.ipynb  processInputImages.py~

Explore-data.ipynb                    __pycache__

Explore-model2.ipynb                  sub-trainModel.py

histogram.png                         trainModel.py

LowRes_13434_overlapping_pairs.h5     trainModel.py~

models                                utilities.py

OverlapSegmentationNet.py             utilities.py~

OverlapSegmentationNet.pyc            utilities.pyc

preprocessing-jp.ipynb                xdata_88x88.npy

processInputImages.py                 ydata_88x88_0123_onehot.npy

In a terminal,  run processInputImages.py as follow:

$python processInputImages.py

The code will generate the two files highlighted in green above.

Training

Switch into gpu mode with (then logout / login):

 

Activate the proper virtual environnement. Mine was called tfgpu, so for my computer, it is:

$source VirtualEnv/tfgpu/bin/activate

Edit the code to set-up the desired number of epoch

From a terminal, run trainModel.py. For example in my computer, it is:

(tfgpu) jeanpat@WA50SHQ:~/image_segmentation_chromosomes/code$ python trainModel.py

If the gpu is properly used by tensorflow the terminal should yield something like:

The usage of the gpu is mentionned, furthermore the cpu usage is not stucked on 100%.

On a modest nvidia GT740M gpu, 14 epochs were run for 8~9 hours.

Using TensorFlow 0.7 from python on a laptop running Manjaro Linux.

The use of GPU version of tensorflow is tested on a laptop running manjaro Linux distribution.

Checking the Nvidia driver installation:

First, check the GPU presence

$ lspci -k | grep -A 2 -E "(VGA|3D)"

00:02.0 VGA compatible controller: Intel Corporation 4th Gen Core Processor Integrated Graphics Controller (rev 06)

        Subsystem: CLEVO/KAPOK Computer Device 5000

        Kernel driver in use: i915

--

01:00.0 3D controller: NVIDIA Corporation GK208M [GeForce GT 740M] (rev ff)

        Kernel modules: nouveau, nvidia

03:00.0 Network controller: Intel Corporation Wireless 7260 (rev 73)

Check if the GPU can be harnessed

 For example with using glxgears. Without GPU, glxgears yields:

$ glxgears

Running synchronized to the vertical refresh.  The framerate should be

approximately the same as the monitor refresh rate.

453 frames in 5.0 seconds = 90.581 FPS

302 frames in 5.0 seconds = 60.288 FPS

XIO:  fatal IO error 11 (Resource temporarily unavailable) on X server ":0"

      after 3146 requests (3146 known processed) with 0 events remaining.

Now let's run glxgears with the gpu acceleration:

$ optirun glxgears

9891 frames in 5.0 seconds = 1977.995 FPS

10186 frames in 5.0 seconds = 2037.131 FPS

9953 frames in 5.0 seconds = 1990.527 FPS

9908 frames in 5.0 seconds = 1981.540 FPS

10129 frames in 5.0 seconds = 2025.626 FPS

glxgears is accelerated by a factor 20 at least.

Installing CUDA 7.5 and cuDNN

CUDA 7.5  was installed from manjaro repository using the Octopi GUI. cuDNN was installed using Octopi. This library has to be first downloaded from the nvidia developer website.

In manjaro Linux, CUDA is installed in /opt/cuda/. To check cuda installation, samples files were built under user directory as explained from Nvidia documentation.

Copy the directory /opt/cuda/samples into your home dir ~/HOME, example:

cp -r /opt/cuda/samples/ ~jeanpat/

Build (with make or make -j4  with a 4 cores CPU ) the test applications in the /samples directory:

$ ls -l
total 144
drwxr-xr-x 48 jeanpat users  4096 11 mai   12:10 0_Simple
drwxr-xr-x  7 jeanpat users  4096 11 mai   12:10 1_Utilities
drwxr-xr-x 12 jeanpat users  4096 11 mai   12:10 2_Graphics
drwxr-xr-x 21 jeanpat users  4096 11 mai   12:10 3_Imaging
drwxr-xr-x 10 jeanpat users  4096 11 mai   12:10 4_Finance
drwxr-xr-x 10 jeanpat users  4096 11 mai   12:10 5_Simulations
drwxr-xr-x 31 jeanpat users  4096 11 mai   12:10 6_Advanced
drwxr-xr-x 37 jeanpat users  4096 11 mai   12:10 7_CUDALibraries
drwxr-xr-x  6 jeanpat users  4096 11 mai   12:10 common
-rw-r--r--  1 jeanpat users 96407 11 mai   12:10 EULA.txt
-rw-r--r--  1 jeanpat users  2652 11 mai   12:10 Makefile

The binary file deviceQuery can be launched using optirun:

$ optirun ./deviceQuery
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "GeForce GT 740M"
  CUDA Driver Version / Runtime Version          7.5 / 7.5
  CUDA Capability Major/Minor version number:    3.5
  Total amount of global memory:                 2048 MBytes (2147352576 bytes)
  ( 2) Multiprocessors, (192) CUDA Cores/MP:     384 CUDA Cores
  GPU Max Clock rate:                            1032 MHz (1.03 GHz)
  Memory Clock rate:                             900 Mhz
  Memory Bus Width:                              64-bit
  L2 Cache Size:                                 524288 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096)
  Maximum Layered 1D Texture Size, (num) layers  1D=(16384), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(16384, 16384), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 1 copy engine(s)
  Run time limit on kernels:                     Yes
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 1 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 7.5, CUDA Runtime Version = 7.5, NumDevs = 1, Device0 = GeForce GT 740M
Result = PASS

Running tensorflow 0.7 from python

Tensorflow was installed (version with GPU acceleration) according to the documentation in a virtual environment. Then the latest version can be installed using:

pip install --upgrade tensorflow

According to the documentation, the tensorflow installation can be checked as follow:

$ python
...
>>> import tensorflow as tf
>>> hello = tf.constant('Hello, TensorFlow!')
>>> sess = tf.Session()
>>> print(sess.run(hello))
Hello, TensorFlow!
>>> a = tf.constant(10)
>>> b = tf.constant(32)
>>> print(sess.run(a + b))
42
>>>

However, to run properly the python interpreter must be run with optirun:

$ optirun python
Python 2.7.11 (default, Mar  3 2016, 11:00:04)
[GCC 5.3.0] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcudnn.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:105] successfully opened CUDA library libcurand.so locally
>>> hello = tf.constant('Hello, TensorFlow!')
>>> sess = tf.Session()
I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:900] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
I tensorflow/core/common_runtime/gpu/gpu_init.cc:102] Found device 0 with properties:
name: GeForce GT 740M
major: 3 minor: 5 memoryClockRate (GHz) 1.0325
pciBusID 0000:01:00.0
Total memory: 2.00GiB
Free memory: 1.97GiB
I tensorflow/core/common_runtime/gpu/gpu_init.cc:126] DMA: 0
I tensorflow/core/common_runtime/gpu/gpu_init.cc:136] 0:   Y
I tensorflow/core/common_runtime/gpu/gpu_device.cc:717] Creating TensorFlow device (/gpu:0) -> (device: 0, name: GeForce GT 740M, pci bus id: 0000:01:00.0)
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 1.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 2.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 4.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 8.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 16.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 32.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 64.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 128.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 256.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 512.0KiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 1.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 2.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 4.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 8.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 16.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 32.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 64.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 128.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 256.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 512.00MiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 1.00GiB
I tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:51] Creating bin of max chunk size 2.00GiB
>>> print sess.run(hello)
Hello, TensorFlow!
>>> a = tf.constant(10)
>>> b = tf.constant(32)
>>> print sess.run(a+b)
42
>>> 

 Conclusion

At first sight, the issue of the use of CUDA 7.5 with TensorFlow is resolved with TensorFlow 0.7.