Merge pull request #41 from urakubo/210906

210906

README.ja.md

@@ -43,24 +43,22 @@
 ## 動作条件：
 OS：　Microsoft Windows 10 (64 bit) または Linux (Ubuntu 18.04にて動作確認済) 
 
-推奨：NVIDIA社の高性能GPUを搭載したグラフィックスカード (Compute capability 3.5 以上のGPU搭載のもの。GeForce GTX 1080tiなど)。
+推奨：NVIDIA社の高性能GPUを搭載したグラフィックスカード (Compute capability 3.5 以上のGPU搭載のもの。GeForce GTX1080ti, RTX2080ti, RTX3090など)。
 
 - https://developer.nvidia.com/cuda-gpus
 
-注意：現在のUNI-EMはNVIDIA社の最新GPU (RTX30X0, A100など）で動作しません。UNI-EMがTensorflow1.Xに基づいて開発されている一方、最新GPUはTensorflow2.Xに対応しますが、Tensorflow1.Xに対応しないためです。詳細は、下記サイトをご参照ください。
-
-- https://www.pugetsystems.com/labs/hpc/How-To-Install-TensorFlow-1-15-for-NVIDIA-RTX30-GPUs-without-docker-or-CUDA-install-2005/
-- https://qiita.com/tanreinama/items/6fc3c71f21d64e61e006
-
 ## インストール方法：
 Pythonのインストールの必要のないPyinstaller版とPythonソースコードの両方を提供します。
 
 ### Pyinstaller版 (Windows10のみ)：
 1. GPU 版とCPU版を用意しました。いずれかをダウンロードして展開してください。
 
-- Version 0.90.4 (2021/05/31):
-	- [CPU version (Ver0.90.4; 363MB)](https://bit.ly/3uwKHkB)
-	- [GPU version (Ver0.90.4: 1,068 MB)](https://bit.ly/2QWfFFb)
+	- Version 0.92 (2021/09/13; NVIDIA Ampere (RTX30X等) 対応):
+		- [CPU & GPU version (Ver0.92; 2,166 MB)](https://bit.ly/2VFvaDS)
+
+	- 前Version 0.90.4 (2021/05/31; NVIDIA Ampere (RTX30X等) 非対応):
+		- [CPU version (Ver0.90.4; 363 MB)](https://bit.ly/3uwKHkB)
+		- [GPU version (Ver0.90.4: 1,068 MB)](https://bit.ly/2QWfFFb)
 
 2. 公開サンプルデータkasthuri15をダウンロードして適当なフォルダに展開してください。
 	- https://www.dropbox.com/s/pxds28wdckmnpe8/ac3x75.zip?dl=0
@@ -70,15 +68,30 @@ Pythonのインストールの必要のないPyinstaller版とPythonソースコ
 
 4. 上端のドロップダウンメニューより一番左のDojo → Open Dojo Folderを選択して、ダイアログよりkasthuri15フォルダ下のmojoを指定してください。サンプルデータがダウンロードされてDojoが起動します。
 
+* Tensorflow training時、inference時に下のエラーが出る場合は、NVIDIA Driverをアップデートしてください。
+	- tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version
+	- https://www.nvidia.com/Download/index.aspx
+
+* TF1のFFNモデルと、TF2のFFNモデルは同一ではありません。TF1でFFNトレーニングを行ったあと、TF2で更なるトレーニングを行ったり、推論を行う事はできません。
+
+* Tensorflow トレーニングの際に、下の警告が出ることを浦久保は認識しておりますが、解決できておりません。どなたかのご助力をお願いいたします。
+	- WARNING:tensorflow:It seems that global step (tf.train.get_global_step) has not been increased. Current value (could be stable): 0 vs previous value: 0. You could increase the global step by passing tf.train.get_global_step() to Optimizer.apply_gradients or Optimizer.minimize.
+
+
 ### Python版：
-1. Windows10またはLinux (Ubuntu18.04 にて動作確認済)にて Python3.5-3. をインストールしてください。
-2. Tensorflow 1.14 のためにGPUを利用する場合はcuda 10.0 および cuDNN 7.4 をインスト―ルしてください **[参考1]** 。
+1. Windows10またはLinux (Ubuntu18.04 にて動作確認済)にて Python3.6- をインストールしてください。
+2. Tensorflow 1.15, Tensorflow2.0- にて動作します。 NVIDIA GPU 利用する場合、Tensorflow 2.4.1の場合は"CUDA 11.0, cuDNN 8.0.4", TensorFlow 2.5.0の場合は"CUDA 11.2.2, cuDNN 8.1.1"をインスト―ルしてください **[参考1]** 。
 3. 次の命令を実行してGithubより必要プログラムをダウンロードしてください。
 
 	- git clone https://github.com/urakubo/UNI-EM.git
 
 
-4. requirements-[os]-[cpu or gpu].txtを参考に、Pythonに必要モジュール「Tensorflow-gpu 1.12, PyQt5, openCV3, pypng, tornado, pillow, libtiff, mahotas, h5py, lxml, numpy, scipy, scikit-image, pypiwin32, numpy-stl」を pip install -r requirements-[os]-[cpu or gpu].txt などのコマンドを用いてインストールしてください。
+4. requirements-[os].txtを参考に、Pythonに必要モジュールを pip install -r requirements-[os].txt などのコマンドによりインストールしてください。Ubuntu18.04, Ubuntu20.04の場合は、 opencv, pyqt5 は下のコマンド "apt" でインストールしてください。
+
+	- sudo apt install python3-dev python3-pip
+	- sudo apt install python3-opencv
+	- sudo apt install python3-pyqt5
+	- sudo apt install python3-pyqt5.qtwebengine
 
 5. コマンドププロンプトにて[UNI-EM]フォルダへ移動して、 python main.py と実行してコントロールパネルを起動してください。
 6. 公開サンプルデータkasthuri15をダウンロードして適当なフォルダに展開してください。
@@ -87,6 +100,19 @@ Pythonのインストールの必要のないPyinstaller版とPythonソースコ
 
 7. 上端のドロップダウンメニューより一番左のDojo → Open Dojo Folderを選択して、ダイアログよりkasthuri15フォルダ下のmojoを指定してください。サンプルデータがダウンロードされてDojoが起動します。
 
+* Tensorflow training時、inference時に下のエラーが出る場合は、NVIDIA Driverをアップデートしてください。
+	- tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version
+	- https://www.nvidia.com/Download/index.aspx
+
+* TF1のFFNモデルと、TF2のFFNモデルは同一ではありません。TF1でFFNトレーニングを行ったあと、TF2で更なるトレーニングを行ったり、推論を行う事はできません。
+
+* Windows10の場合、下のTF1.15.4 と cuda 11.1 の組み合わせで、1.4倍程度トレーニング速度が速くなることを確認しています。 
+	- https://github.com/fo40225/tensorflow-windows-wheel/tree/master/1.15.4+nv20.12/
+
+* Tensorflow トレーニングの際に、下の警告が出ることを浦久保は認識しておりますが、解決できておりません。どなたかのご助力をお願いいたします。
+	- WARNING:tensorflow:It seems that global step (tf.train.get_global_step) has not been increased. Current value (could be stable): 0 vs previous value: 0. You could increase the global step by passing tf.train.get_global_step() to Optimizer.apply_gradients or Optimizer.minimize.
+
+
 ## お願い：
 日本国内の実験研究者、情報学研究者さまのフィードバックをお待ちします（hurakubo あっと gmail.com; **[参考3]** ）。私一人で開発を続けることは困難なので、共同開発者も募集いたします。本アプリは、自然画像のセグメンテーション等に利用することも可能と思われますので、多様なコメントをお待ちしております。本アプリの開発には、革新脳、新学術、基盤Cのご支援をいただいております。
 

README.md

@@ -42,32 +42,25 @@ Multiple users can simultaneously use it through web browsers. The goal is to de
 ## System requirements
 Operating system: Microsoft Windows 10 (64 bit) or Linux (Ubuntu 18.04).
 
-Recommendation: High-performance NVIDIA graphics card whose GPU has over 3.5 compute capability.
+Recommendation: High-performance NVIDIA graphics card whose GPU has over 3.5 compute capability (e.g., GeForce GTX1080ti, RTX2080ti, and RTX3090).
 
 - https://developer.nvidia.com/cuda-gpus
 
-Caution: currently, UNI-EM cannot run on the newest NVIDIA GPUs, such as A100 and RTX30X0 (in particular, under microsoft windows). This is because UNI-EM is based on tensorflow1.X, while the newest GPUs are compatible with tensorflow2.X, but not tensorflow1.X. Please refer to the following website.
-
-- https://www.pugetsystems.com/labs/hpc/How-To-Install-TensorFlow-1-15-for-NVIDIA-RTX30-GPUs-without-docker-or-CUDA-install-2005/
-- https://developer.nvidia.com/blog/accelerating-tensorflow-on-a100-gpus/
-- https://github.com/NVIDIA/tensorflow
-
 ## Installation
 We provide standalone versions (pyinstaller version) and Python source codes.
 
 ### Pyinstaller version (Microsoft Windows 10 only)
 1.	Download one of the following two versions, and unzip it:
 
-- Version 0.90.4 (2021/05/31):
-	- [CPU version (Ver0.90.4; 363MB)](https://bit.ly/3uwKHkB)
-	- [GPU version (Ver0.90.4: 1,068 MB)](https://bit.ly/2QWfFFb)
+- Version 0.92 (2021/09/13):
+	- [CPU & GPU version (Ver0.92; 2,166 MB)](https://bit.ly/2VFvaDS)
 
 - Release summary:
-	- Bug fix.
-		- Bug fix version of FFNs was used.
-		- Tentative solution in “Cannot lock file” error in the inference of 2D CNN.
-		- Safe launch of Tensorboard.
-		- Abolish the use of mcube (caused an occasional error in launching).
+	- Compatibility with both tensorflow1.X and 2.X. It now works also on NVIDIA Ampere GPUS (RTX30X0, etc).
+		- Caution: FFN model in TF2 is not identical to that in TF1. Trained model using TF1 cannot be used for further training or inference in TF2. 
+	- Revision of documents for FFNs.
+	- Bug fix (Tensorboard, 2D/3D watersheds, Filetypes of 2D CNN inference, etc).		
+
 
 2.	Download one of sample EM/segmentation dojo folders from the following link, and unzip it:
    	- https://www.dropbox.com/s/pxds28wdckmnpe8/ac3x75.zip?dl=0
@@ -77,9 +70,19 @@ We provide standalone versions (pyinstaller version) and Python source codes.
 
 4.	Select Dojo → Open Dojo Folder from the dropdown menu, and specify the folder of the sample EM/segmentation dojo files.  The proofreading software Dojo will be launched.
 
+* Update the dirver of NVIDIA GPU if you see the following error.
+	- tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version
+	- https://www.nvidia.com/Download/index.aspx
+
+* Caution: FFN model in TF1 is not consistent with that in TF2. The trained model using TF1 cannot be served for the further training or inference in TF2.
+
+* In the process of traning, HU sees the following warning, and has not found out how to suppress it. I ask someone for help.
+	- WARNING:tensorflow:It seems that global step (tf.train.get_global_step) has not been increased. Current value (could be stable): 0 vs previous value: 0. You could increase the global step by passing tf.train.get_global_step() to Optimizer.apply_gradients or Optimizer.minimize.
+
+
 ### Python version 
-1. Install Python 3.5-3.7 in a Microsoft Windows PC (64 bit) or Linux PC (Ubuntu 18.04 confirmed).
-2. Install cuda 10.0 and cuDNN 7.4 for Tensorflow 1.14 if your PC has a NVIDIA-GPU.
+1. Install Python 3.6- in a Microsoft Windows PC (64 bit) or Linux PC (Ubuntu 18.04 confirmed).
+2. Install "cuda 11.0 and cuDNN 8.0.4 for Tensorflow 2.4.1", or "cuda 11.2.2 and cuDNN 8.1.1 for Tensorflow 2.5.0" if your PC has a NVIDIA-GPU.
 
 	- https://www.tensorflow.org/install/source
 	- https://www.tensorflow.org/install/source_windows
@@ -88,9 +91,12 @@ We provide standalone versions (pyinstaller version) and Python source codes.
 
 	- git clone https://github.com/urakubo/UNI-EM.git
 
-4. Install the following modules of Python: Tensorflow-gpu, PyQt5, openCV3, pypng, tornado, pillow, libtiff, mahotas, h5py, lxml, numpy, scipy, scikit-image, pypiwin32, numpy-stl. Check "requirements-[os]-[cpu or gpu].txt". Users can install those module using the following command.
+4. Install the following modules of Python: Tensorflow-gpu, PyQt5, openCV3, pypng, tornado, pillow, libtiff, mahotas, h5py, lxml, numpy, scipy, scikit-image, pypiwin32, numpy-stl. Use the command "requirements-[os]-.txt". Use the following "apt" commands to install opencv and pyqt5 if you use Ubuntu/Linux:
 
-	- pip install -r requirements-[os]-[cpu or gpu].txt
+	- sudo apt install python3-dev python3-pip
+	- sudo apt install python3-opencv
+	- sudo apt install python3-pyqt5
+	- sudo apt install python3-pyqt5.qtwebengine
 
 5. Download one of sample EM/segmentation dojo folders from the following link, and unzip it:
    	- https://www.dropbox.com/s/pxds28wdckmnpe8/ac3x75.zip?dl=0
@@ -100,6 +106,15 @@ We provide standalone versions (pyinstaller version) and Python source codes.
 
 7.	Select Dojo → Open Dojo Folder from the dropdown menu, and specify the sample EM/segmentation dojo folder. The proofreading software Dojo will be launched.
 
+* Update the dirver of NVIDIA GPU if you see the following error.
+	- tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version
+	- https://www.nvidia.com/Download/index.aspx
+
+* Caution: FFN model in TF1 is not consistent with that in TF2. The trained model using TF1 cannot be served for the further training or inference in TF2.
+
+* In the process of traning, HU sees the following warning, and has not found out how to suppress it. I ask someone for help.
+	- WARNING:tensorflow:It seems that global step (tf.train.get_global_step) has not been increased. Current value (could be stable): 0 vs previous value: 0. You could increase the global step by passing tf.train.get_global_step() to Optimizer.apply_gradients or Optimizer.minimize.
+
 ## Authors
 
 * [**Hidetoshi Urakubo**](https://researchmap.jp/urakubo/?lang=english) - *Initial work* - 

docs/HowToUse.md

@@ -114,39 +114,36 @@ The VAST Lite is recommended for 3D ground truth generation (https://software.rc
 
 #### Procedure:
 1.	Select Segmentation → 3D FFN in the pulldown menu. A dialog that has the four tabs appears: Preprocessing, Training, Inference, and Postprocessing.
-2.	Select the preprocessing tab and specify parameters:
-	- Image Folder:	Folder containing EM images (grayscale sequential tiff/png images).
-	- Ground Truth Folder: Folder containing ground truth segmentation (grayscale sequential tiff/png images).
-	- FFN File Folder: 	Folder storing generated files for training.
+2.	Select the preprocessing tab and specify folders:
+	- Training Image Folder: Folder containing EM images for training (sequentially numbered grayscale tiff/png images).
+	- Ground Truth Folder: Folder containing ground truth segmentation (sequentially numbered grayscale tiff/png images).
+	- Empty Folder for FFNs: Empty folder to store generated preprocessed files for training.
 	- Save Parameters
 	- Load Parameters
 
-	Users will see an example EM image volume and their segmentation (kasthuri15) by downloading the following example data.
-	- ExampleFFN.zip 522MB: https://www.dropbox.com/s/cztcf8w0ywj1pmz/ExampleFFN.zip?dl=0
+	Users can use an example EM image volume and their segmentation (kasthuri15) by downloading the following example data.
+	- ExampleFFN.zip 154MB: https://www.dropbox.com/s/06eyzakq9o87cmk/ExampleFFN.zip?dl=0
 
-3.	Execute the preprocessing. It takes 5 to 60 min depending on target image volume and machine speed. It produces three files in the FFN file folder: af.h5, groundtruth.h5, and tf_record_file .
+3.	Execute the preprocessing. It takes 5 to 60 min depending on target image volume and machine speed. It produces three files in the Empty Folder for FFNs: af.h5, groundtruth.h5, and tf_record_file .
 4.	Select the training tab and specify parameters:
 	- Max Training Steps: 	The number of training FFN, a key parameter.
 	- Sparse Z:	Check it if the target EM-image stack is anisotropic.
-	- Training Image h5 File:	Generated file
-	- Ground truth h5 File:		Generated file.
-	- Tensorflow Record File:	Generated file.
-	- Tensorflow Model Folder:	Folder storing training results.
+	- FFNs Folder:  Folder storing generated preprocessed files.
+	- Model Folder: Folder to store trained Tensorflow model.
 5.	Execute the training. It requires over a few days depending on the target image volume, machine speed, and the Max Training Steps. A few million training steps are required for minimal quality inference. Users can execute additive training by specifying the same parameter settings with the increasing number of "Max Training Steps".
-6.	Select the inference tab and specify parameters:
-	- Target Image Folder:	Folder containing EM images (sequential grayscale tiff/png images).
-	- Output Inference Folder: Folder that will store the inference result.
-	- Tensorflow Model Files: Specify the trained model files. Please remove their suffix, and just specify the prefix such as "model.ckpt-2000000."
+6.	Select the inference tab and specify folders:
+	- Target Image Folder:	Folder containing EM images for inference (sequentially numbered grayscale tiff/png images).
+	- Model Folder: Folder storing the trained Tensorflow model, i.e., trios of ”model.ckpt-XXXXX.data-00000-of-00001", "model.ckpt-XXXXX.index", and "model.ckpt-4000000.meta".
+	- FFNs Folder:  FFNs Folder. Inference results will be stored in this folder.
 	- Sparse Z:	Check if it was checked it at the training process.
 	- Checkpoint interval: Checkpoint interval.
-	- FFN File Folder: Folder storing generated files for inference "inference_params.pbtxt."
 	- Save Parameters
 	- Load Parameters
-7.  Execute the inference. It requires 5-60 min depending on target image volume and machine speed. It produces inference results "0/0/seg-0_0_0.npz " and " seg-0_0_0.prob " in the Output Inference Folder. It also produces "inference_params.pbtxt" in the FFN file folder.
+7.  Execute the inference. It requires 5-60 min depending on target image volume and machine speed. It produces inference results " 0/0/seg-0_0_0.npz " and " 0/0/seg-0_0_0.prob " in the FFNs Folder. It also produces "inference_params.pbtxt" in the FFNs folder.
 8.  Select the postprocessing tab and specify parameters:
-	- Target Inference File: Specify inferred segmentation file such as seg-0_0_0.npz.
-	- Output Inference Folder: Folder storing generated sequential image files.
-	- OUtput Filetype: Please select one of them. 16 bit images are recommended.
+	- FFNs Folder:  FFNs Folder that stores inference results, i.e., 0/0/seg-0_0_0.npz.
+	- Output Segmentation Folder (Empty): Folder to store generated sequential segmentation images.
+	- Output Filetype: Select one of them. 8 bit color PNG is good for visual inspection. 16 bit gray-scale filetypes are good for further analyses.
 	- Save Parameters
 	- Load Parameters
 9.  Execute the postprocessing. It generally requires less than 5 min. It produces the inference result in the Output Inference Folder.