Proper Masking in Product Detection
13 Questions
0 Views

Proper Masking in Product Detection

Created by
@SprightlyRetinalite28

Questions and Answers

When should a mask be drawn around the entire empty space on the shelf?

When possible to put at least one product

Why should you mask only visible SKUs in a probe?

  • To focus on less important details
  • To ignore unclear products (correct)
  • To speed up the task
  • To minimize the masking effort
  • QATs should mask all the POSMs regardless of the products they contain.

    False

    What is Masking?

    <p>Drawing rectangle or squares to cover the full products, along the border in order to identify separate products by the system.</p> Signup and view all the answers

    Why is Proper Masking Important?

    <p>Proper masking is important to the success of the process as it affects the engine's training and accuracy, damages voting accuracy, leads to wrong PPA calculation, and can hold back pending developments.</p> Signup and view all the answers

    What is the purpose of the feature Re-masking?

    <p>To enable sending probes for re-masking without losing current labeled data.</p> Signup and view all the answers

    What should Maskers always do when masking the products?

    <p>Mask all the visible products in the probe</p> Signup and view all the answers

    Transparent packs should be masked individually. (True/False)

    <p>True</p> Signup and view all the answers

    What category is one of the main KPI for Tesco and Aldi?

    <p>Fresh recognition</p> Signup and view all the answers

    What is done before creating a masking task in SMART?

    <p>Rectify the images</p> Signup and view all the answers

    Unclear Piles (bad ordered) should be masked as ____ ____.

    <p>one unit</p> Signup and view all the answers

    Masker should re-draw all masks if the screen size is changed during the task.

    <p>True</p> Signup and view all the answers

    Match the following masking guidelines with the correct description:

    <p>Cookies &amp; Crackers = Piles of Cookies and Crackers should be masked separately Boxes = Boxes with multiple small products should be masked as one unit Confectionary masking = Small confectionary products should be masked together as one unit Front Products = Only front-facing products should be masked</p> Signup and view all the answers

    Study Notes

    Masking Guide

    • Masking is the process of drawing rectangles or squares to cover full products in a probe to identify separate products in a system.
    • The goal of masking is to use machine-learning methods to automatically detect and label products in a given probe.
    • Masking is necessary because there is no automatic masking algorithm available, so QATs (Quality Assurance Team) manually draw masks around products as well as they can.

    Importance of Proper Masking

    • Proper masking is crucial for the success of the process, as improper masking can lead to invalid votes in voting, harming the engine's training and accuracy, damaging the accuracy of voting, and affecting PPA calculation.
    • Proper masking enables the system to correctly identify and label products, ensuring accurate results.

    Masking Environment

    • Masking is done in a special environment called "Masking" in SMART.
    • This environment provides features such as filling masks with a dark background, resizing masks, dragging masks, and changing mask colors.

    Masking Features

    • The system has a feature called "Duplicate Mask Button" that enables creating duplicate masks by clicking on an icon.
    • The "Resize Button" allows selecting a mask by double-clicking, enabling the drawing of overlapping masks, resizing buttons, and deleting buttons.

    Masking Guidelines

    • All visible products should be masked in the probe, considering generic tagging rules.
    • The mask should be drawn exactly on the SKU border, not too narrow or too wide.
    • Masks should also be drawn on empty spaces.

    Specific Masking Scenarios

    • If products are partially covered, it is preferable to draw masks with a small overlap rather than drawing a smaller mask than the actual product size.
    • For clip strips and hanging products, each SKU should be masked individually.
    • For hoppers, baskets, and packages, one single mask should be drawn for the entire unit.
    • For transparent packs, each individual product should be masked, while for non-transparent packs, the entire unit should be masked.
    • For tilted products, one big mask should be drawn around the group of products.
    • For rotated probes, the probe should be canceled using the "Probe is Rotated" option.

    Masking POSMs (Point of Sale Materials)

    • All POSMs should be masked.
    • POSM masking should be avoided unless the products are covered by a POSM, in which case each product should be masked individually.

    Masking Probes with Quality Issues

    • If the probe is too blurry, the QAT can skip masking and continue to the next probe.
    • If the image is taken from a big distance, making it impossible to recognize SKUs, the QAT can skip masking.
    • Only visible SKUs should be masked in a probe, and unclear products can be ignored.

    Masking Empty Spaces

    • QATs should mask all empty spaces where possible to put at least one product.
    • The mask should be drawn around the entire empty space on the shelf.
    • However, if an empty space is not suitable for holding products, no mask should be drawn.### Masking Guidelines
    • Mask only visible price labels in the probe, do not mask products.
    • Special guidelines from the R&D team may apply.

    Masking Menus

    • Mask menu items one by one, following menu recognition rules.
    • Consider special guidelines from the R&D team, if applicable.

    Bad Masking

    • Common mistake: products without masks at all.
    • Mask quality issues: masks too narrow, not drawn at all, or drawn incorrectly.

    Partial Masking

    • Mask only products with tags, do not mask other products.
    • If a product does not have a tag, a live alert will pop up, and the incorrect masking will disappear.

    Rectification for Masking Tasks

    • Mechanism in SMART rectifies images from Dome cameras to support "Fresh" recognition for Tesco and Aldi.
    • Rectification is done before creating a masking task, and the masker will mask the rectified images.

    Delete Masks When Screen Size is Changed

    • Solution to avoid masking offset issue: delete masks and notify masker with a pop-up message if screen size is changed during the task.

    Special Masking Guidelines

    • Piles Masking: unclear piles (bad ordered) should be masked as one unit.
    • Confectionary Masking: small confectionary products laying on top of each other should be masked once as one unit.
    • Cookies & Crackers: well-ordered piles should be masked separately.
    • FRONT PRODUCTS: only facing/front products should be masked.
    • BOXES: boxes with multiple small products in them should be masked as one unit.

    Full Guided Masking (FGM)

    • Flow: 1st cycle of voting, followed by masking task, and if needed, another cycle of voting.
    • SLA for both voting and masking is 24 hours.
    • Process: maskers can mask according to tagged products without recognition confusion.
    • Full Guided Masking (FGM) – Empty Masking: QATs should mask all empty space separately with one tag each.

    Full Guided Masking (FGM) – Transparent Packs

    • Scenario 01: transparent pack with tags on each product - mask each product with a tag.
    • Scenario 02: transparent pack with one tag on the pack - mask the entire transparent package.

    Mssing Mask Alert – PM & FGM

    • Alert 01: missing masks - not all tags are masked; confirm and continue only after ensuring all relevant tags are masked.
    • Alert 02: missing masks - zero masks were drawn; cancel the probe if only dislocated/unnecessary tags are present, and mention the cancelation reason properly.

    Studying That Suits You

    Use AI to generate personalized quizzes and flashcards to suit your learning preferences.

    Quiz Team

    Description

    Learn about the importance of masking in product detection, including its purpose, manual process, and role in machine learning methods.

    More Quizzes Like This

    Use Quizgecko on...
    Browser
    Browser