Salesvision API

Salesvision API provides accurate, reliable and scalable fashion image analysis by endpoints.

It includes free public methods and python API client

Fashion analysis

Fashion analysis includes recognition of items:

• Category
• Tags/attributes
• Mask
• Color

Category detection

The following solution will detect 27 categories:

Tag/attribute recognition

The solution will specify tags from 7 main categories and styles such as described below:

On the real world images will have such output for different fashion items:

Color extraction

Here is example of how color is been extracted from the fashion item mask covered area:

Public methods

Curl HTTP Request

Public endpoint for fashion analysis with time measurement

`IMG_URL=https%3A%2F%2Fis4.revolveassets.com%2Fimages%2Fp4%2Fn%2Fd%2FCAME-WD114_V1.jpg;
time curl -X GET "http://87.117.25.190:5015/fashion_analysis/?url=$IMG_URL"`

From office local network

`IMG_URL=https%3A%2F%2Fis4.revolveassets.com%2Fimages%2Fp4%2Fn%2Fd%2FCAME-WD114_V1.jpg;
time curl -X GET "http://192.168.0.125:5015/fashion_analysis/?url=$IMG_URL"`

Try it yourself at:

office local network access link

global network access link

Parameter	Default	Description
url	true	fashion image url.

The json output will contain result list where each object is a recognized fashion item

Each object will include:

• 1 of 27 fashion classes
• Tags description in 7 main categories
• Mask of the recognized item
• 5 main colors extracted from item mask covered area
• Color embedding which can be used later on with fashion item search by color

Visual analysis method

The above command returns JSON structured like this:

{
  "result": [
    {
      "category": "pants",
      "description": {
        "tags": {
          "length": "maxi (length)",
          "nickname": "jeans",
          "opening type": "fly (opening)",
          "silhouette": "regular (fit)",
          "textile finishing, manufacturing techniques": "washed",
          "textile pattern": "plain (pattern)",
          "waistline": "low waist"
        },
        "colors": [
          "172839",
          ...
        ]
      },
      "color_embedding": [
        -0.45759817957878113,
        ...
      ]
    },
    {
      "category": "top, t-shirt, sweatshirt",
      "description": {
        "tags": {
          "length": "above-the-hip (length)",
          "nickname": "classic (t-shirt)",
          "opening type": "no opening",
          "silhouette": "symmetrical",
          "textile finishing, manufacturing techniques": "printed",
          "textile pattern": "plain (pattern)",
          "waistline": "no waistline"
        },
        "colors": [
          "321d1a",
          ...
        ]
      },
      "color_embedding": [
        -0.5404209494590759,
        ...
      ]
    },
    {
      "category": "shoe",
      "description": {
        "colors": [
          "161615",
          ...
        ]
      },
      "color_embedding": [
        -0.5041476488113403,
        ...
      ]
    },
    {
      "category": "headband, head covering, hair accessory",
      "description": {
        "colors": [
          "35261b",
          ...
        ]
      },
      "color_embedding": [
        -0.5759932398796082,
        ...
      ]
    }
  ]
}

That method allows us to check accuracy of given masks and bounding boxes recognition by the given image:

Provided attributes for the recognized "category": "pants":

Attribute	Predicted tag
length	maxi
nickname	jeans
opening type	fly
silhouette	regular (fit)
textile finishing, manufacturing techniques	washed
textile pattern	plain
waistline	low waist

Try it yourself at:

office local network access link

global network access link

Client API

Setup

You can easily setup our SDK with python 3.x language

Install pip package (under development)

pip install salesvision

Authentication

Example of authentification process: will be replaced with fastapi oauth2

from salesvision import Salesvision

# connect to Salesvision module to work with its API

api = Salesvision(api_url='https://salesvision.com/api', api_key='Your_API_key', secret)

Make sure to replace v with your API key.

Salesvision API will be probably using oAuth2 for authentification process

Authorization: Your_API_key

Fashion analysis

from salesvision import Salesvision

# connect to Salesvision module to work with its API
api = Salesvision(api_url='https://salesvision.com/api', api_key='Your_API_key', secret)


# image can be rather url or local stored file
results = api.fashion_analysis(image)

The above command returns JSON structured like this:

{
  "result": [
    {
      "category": "pants",
      "description": {
        "tags": {
          "length": "maxi (length)",
          "nickname": "jeans",
          "opening type": "fly (opening)",
          "silhouette": "regular (fit)",
          "textile finishing, manufacturing techniques": "washed",
          "textile pattern": "plain (pattern)",
          "waistline": "low waist"
        },
        "colors": [
          "172839",
          ...
        ]
      },
      "color_embedding": [
        -0.45759817957878113,
        ...
      ]
    },
    {
      "category": "top, t-shirt, sweatshirt",
      "description": {
        "tags": {
          "length": "above-the-hip (length)",
          "nickname": "classic (t-shirt)",
          "opening type": "no opening",
          "silhouette": "symmetrical",
          "textile finishing, manufacturing techniques": "printed",
          "textile pattern": "plain (pattern)",
          "waistline": "no waistline"
        },
        "colors": [
          "321d1a",
          ...
        ]
      },
      "color_embedding": [
        -0.5404209494590759,
        ...
      ]
    },
    {
      "category": "shoe",
      "description": {
        "colors": [
          "161615",
          ...
        ]
      },
      "color_embedding": [
        -0.5041476488113403,
        ...
      ]
    },
    {
      "category": "headband, head covering, hair accessory",
      "description": {
        "colors": [
          "35261b",
          ...
        ]
      },
      "color_embedding": [
        -0.5759932398796082,
        ...
      ]
    }
  ]
}

Under the hood POST request is used

Query Parameters

Parameter	Default	Description
file:	false	file in binary format.

The output of this method will contain result list where each object is a recognized fashion item

Each object will include:

• 1 of 27 fashion classes
• Tags description in 7 main categories
• Mask of the recognized item
• 5 main colors extracted from item mask covered area
• Color embedding which can be used later on with fashion item search by color