Purpose: This document explains how Sapien would like taggers to tag within the Vehicle Positioning task. The overall goal of the Vehicle Positioning dataset is to train AI models to recognize vehicles accurately by tagging each image based on its angle and view.

Vehicle image labeling task

Given dataset consists of images of vehicles from various angles, both interior and exterior. Each image needs to be labeled with an integer number according to the following point system, based on how the vehicle is photographed:

• Front angle:
  • Vehicle must be photographed from the front at approximately 45 degrees
  • Entire vehicle myst be visible in the frame
  • Both front and side of vehicle should be clearly visible
• Front:
  • Vehicle photographed from a direct front-facing position
  • Can be slightly angled is the vehicle does not fit fully in the frame
• Side:
  • Complete side profile view of vehicle facing left or right
  • Should show full length of vehicle from side to side
• Back:
  • Rear view of vehicle, either straight on or at an angle < 45 degrees
  • Focus is on rear view
• Interior / close-up:
  • Any interior shots of the vehicle
  • Close-ups of specific vehicle components
  • Images that don’t fit other categories

The point categories apply both for undamaged and damaged vehicles. Please see the examples given below.

Examples:

Front angle.  Front facing at an angle of around 45 degrees

Front. Facing straight forward or forward at an angle if vehicle does not fully fit

Side. Facing to the side

Back. Facing straight back or back at an angle

Interior / close-up. Interior images, very tight close up shots, images of the engine, open doors, wheels etc.