Wolfram Computation Meets Knowledge

Inception V1 Trained on Extended Salient Object Subitizing Data

Count the number of prominent items in an image

Released in 2016, this neural net is a fine-tuning of the Inception V1 model trained on the Extended Salient Object Subitizing dataset. In order to improve the performance, the dataset has been enlarged by more than a factor of 2 with synthetic images, obtained by placing segmented objects over background images.

Number of layers: 147 | Parameter count: 5,978,677 | Trained size: 24 MB

Training Set Information

Examples

Resource retrieval

Retrieve the resource object:

In[1]:=
ResourceObject["Inception V1 Trained on Extended Salient Object \
Subitizing Data"]
Out[1]=

Get the pre-trained net:

In[2]:=
NetModel["Inception V1 Trained on Extended Salient Object Subitizing \
Data"]
Out[2]=

Basic usage

Obtain the number of salient objects in an image:

In[3]:=
Out[3]=

Obtain the probabilities:

In[4]:=
Out[4]=

Export to MXNet

Export the net into a format that can be opened in MXNet:

In[5]:=
jsonPath = 
 Export[FileNameJoin[{$TemporaryDirectory, "net.json"}], 
  NetModel["Inception V1 Trained on Extended Salient Object \
Subitizing Data"], "MXNet"]
Out[5]=

Export also creates a net.params file containing parameters:

In[6]:=
paramPath = FileNameJoin[{DirectoryName[jsonPath], "net.params"}]
Out[6]=

Get the size of the parameter file:

In[7]:=
FileByteCount[paramPath]
Out[7]=

The size is similar to the byte count of the resource object:

In[8]:=
ResourceObject[
  "Inception V1 Trained on Extended Salient Object Subitizing \
Data"]["ByteCount"]
Out[8]=

Represent the MXNet net as a graph:

In[9]:=
Import[jsonPath, {"MXNet", "NodeGraphPlot"}]
Out[9]=

Requirements

Wolfram Language 11.2 (September 2017) or above

Reference