# 2. How to Use ReNomIMG GUI tool ¶

## 2.1. Start the Application ¶

ReNomIMG is a single page web application. If your installation have done successfully, you can run application in any directory with following commands.

cd workspace # Workspace can be any directory in your pc.
renom_img # This command will starts ReNomIMG GUI server.


For the command  renom_img  , you can give following arguments.

• –host : This specifies server address.
• –port : This specifies port number of the server.

For example, following code runs ReNomIMG with port 8888.

renom_img --port 8888 # Running ReNomIMG with port 8888


If the application server runs, open web browser and type the server address to the address bar like this.

Then the application will be appeared.

## 2.2. Place your dataset ¶

When the server starts,  datasrc  directory and  storage  directory will be created in the server running directory.

The  datasrc  directory has following folder structure.

datasrc/
├── img   # Set training img files here.
├── label
│   ├── classification # Set classification training label files here
│   ├── detection      # Set detection training label files here
│   └── segmentation   # Set segmentation training label files here
└── prediction_set
├── img     # Set prediction img files here.
└── output  # Prediction result will be output here.
├── classofication # Set classofication training label files here.
├── detection      # Set detection training label files here.
│     ├── csv
│     └── xml
└── segmentation   # Set segmentation training label files here.


As written in the above comments, please set training image data to  datasrc/img  , set training label data to  datasrc/label  .

Note

The name of image file and corresponded label file name have to be same. For example, the image file name is  image01.jpg  , corresponded label file name have to be  image01.xml  .

### 2.2.1. Format of the Detection data ¶

Format of image files : ReNomIMG only accepts  JPEG  and  PNG  formatted image files.

Format of label files : ReNomIMG only accepts  xml  formatted label files. The format of xml file is bellow.

put xml files here :  <ReNomIMG dir>datasrc/label/detection/<sample.xml> 

<annotation>
<size>
<width>374</width>
<height>500</height>
<depth>3</depth>
</size>
<object>
<name>car</name>
<bndbox>
<xmin>2</xmin>
<ymin>3</ymin>
<xmax>374</xmax>
<ymax>500</ymax>
</bndbox>
</object>
</annotation>


ReNomIMG accepts PASCAL VOC formatted object detection data.

The PASCAL Visual Object Classes

### 2.2.2. Format of the Classification data ¶

Format of label files : ReNomIMG only accepts  txt  formatted label files. The format of text file is bellow.

Please Save as  target.txt 

put here :  <ReNomIMG dir>datasrc/label/classification/target.txt 

crayfish_image_0035.jpg crayfish
crayfish_image_0065.jpg crayfish
crayfish_image_0037.jpg crayfish
crayfish_image_0032.jpg crayfish
crayfish_image_0028.jpg crayfish
crayfish_image_0051.jpg crayfish
wrench_image_0035.jpg wrench
wrench_image_0037.jpg wrench
wrench_image_0032.jpg wrench
wrench_image_0028.jpg wrench
wrench_image_0019.jpg wrench
wrench_image_0031.jpg wrench
...
...
...
...
...
pigeon_image_0035.jpg pigeon
pigeon_image_0037.jpg pigeon
pigeon_image_0032.jpg pigeon
pigeon_image_0028.jpg pigeon
pigeon_image_0019.jpg pigeon
pigeon_image_0031.jpg pigeon
pigeon_image_0012.jpg pigeon
pigeon_image_0002.jpg pigeon
pigeon_image_0015.jpg pigeon
pigeon_image_0042.jpg pigeon
pigeon_image_0036.jpg pigeon
pigeon_image_0022.jpg pigeon
pigeon_image_0021.jpg pigeon
pigeon_image_0029.jpg pigeon


ReNomIMG accepts PASCAL VOC formatted object detection data.

The PASCAL Visual Object Classes

### 2.2.3. Format of the Segmentation data ¶

Warning

Segmentation require two kind of labels.  PNG  files and  class_map.txt 

Format of image files : ReNomIMG only accepts  JPEG  and  PNG  formatted image files.

Format of label files : ReNomIMG only accepts  txt  and  PNG  formatted label files. The format of txt file is bellow.

Please Save as  class_map.txt  .

Put file here :  <ReNomIMG dir>/datasrc/label/segmentation/class_map.txt 

Good example

• Class number id must be start 0 and set background.
• Class number id must be serial number.
background 0
airplane 1
bicycle 2
bird 3
boat 4
bottle 5
bus 6
car 7
cat 8
chair 8
cow 10
diningtable 11
dog 12
horse 13
motorbike 14
person 15
potted plant 16
sheep 17
sofa 18
train 19
tv/monitor 20


• Class number id does not start 0.
• Class name does not have background.
• Class number does not have serial number.
airplane 1
bicycle 10
bird 50
boat 100
bottle 150
bus 200
car 250
cat 300
chair 350
cow 400
diningtable 450
dog 500
horse 550
motorbike 600
person 700
potted plant 750
sheep 800
sofa 900
train 950
tv/monitor 1000


Sample of Segementation PNG label file

ReNomIMG accepts PASCAL VOC formatted object detection data.

The PASCAL Visual Object Classes

Note

The name of image file and corresponded label file name have to be same. For example, the image file name is  image01.jpg  , corresponded label file name have to be  image01.png  .

## 2.3. Create Detection Model ¶

So far, the server and dataset are prepared. Let’s build a object detection model. For building a model, you have to specify  dataset  and  hyper parameters  .

### 2.3.1. Create Dataset ¶

For training a machine learning model, you have to prepare training dataset and validation dataset. Training dataset is used for training model, and validation dataset is used for evaluating a model in terms of how accurately predict data that have not used in training.

In ReNomIMG, training dataset and validation dataset will be randomly sampled from the data that is in the  datasrc  directory.

According to the above figure, you can create  dataset  from datasrc. Once a dataset is created its content will never be change.

For creating a  dataset  , please move to dataset setting modal. Following figures guide you to the dataset page.

Then following page will be appeared.

As you can see, you can specify the  dataset name  ,  description  and  ratio of training data  .

After filling all forms, please push the  confirm  button to confirm the content that the dataset includes.

Then following graph will be appeared. You can confirm what classes are included in the dataset and how many tags are they.

At last, for saving the dataset, please push the  submit  button.

You can confirm created datasets in the dataset page. For going to the dataset page, please follow the figure below.

In the above figure, 2 datasets are already created.

### 2.3.2. Hyper parameter setting ¶

So far you got all the materials, let’s build a model and run training. For creating a model please push the button  New  .

Then you can see a hyper parameter setting modal like following figure.

As you can see in above figure, you can specify following parameters.

• Dataset Name … Select the dataset for training.
• CNN architecture … Select the object detection algorithm.
• Batch Size … Set higher number, the learning speed become fast . Set lower number, the behavior become like Stochastic Gradient Descent(or just SGD)
• Total Eopch … Training your network on each item of the set once is an epoch.
• Train Whole network … If this is checked, whole network weight will be trained.
• Image size … Image size for training.

Note

Depending on your GPU device, larger image size or batch size causes memory overflow.

### 2.3.3. Training Model ¶

Finishing hyper parameter settings, then push run button to start training!

If the training starts, model will be appeared in model list and progress bar will be shown.

Note

Detection, Segmentation and Classification task take same process.

## 2.4. Perform Prediction ¶

After finishing the training, we can use the model for predicting new image data.

In the ‘Train Page’ you can see Deploy button. The deployed model will be used for performing prediction. If any model is deployed you will see following view.

Then please move to ‘Predict Page’ using the side bar menu. The following picture is a ‘Prediction Page’.

You will see Run Prediction button. Pushing this button runs the prediction using deployed model.

Note

The input image of the prediction will be the images that aligned to datasrc/prediction_set/img . Required directory structure is described in Place your dataset .

After the prediction you will see the result on the screen like following picture.

Also you can download the result as csv file. The button placed on the right top of the prediction result allows you to download the result.

## 2.5. Uninstall ReNomIMG ¶

You can uninstall ReNomIMG by following pip command.

pip uninstall renom_img