This page gives a more detailed overview of Hive's visual moderation classes related to drugs, smoking, etc. If you need more details on these classes after reading our main Visual Moderation page, look here. We'll enumerate, as clearly as possible, which types of subject matter are covered by each class in our models.
Because all platforms have different moderation requirements and risk sensitivities, we recommend that you consult these descriptions carefully as you decide which classes to build into your moderation logic. At the end of the day, it's up to you to decide which classes are important to monitor based on your content policies.
To determine which class(es) cover specific types of visual content, it may be helpful to search this page (Ctrl/Cmd + F) with terms for that subject matter (i.e., gun, injury) rather than looking for it in specific class descriptions.
Before looking at subject matter breakdowns for each class, it may be helpful to understand the following:
Hive's visual classifier is multi-headed. Each model head defines a group of categorizations we call classes. Each model head includes at least one positive class (e.g., yes_smoking) and a negative class (e.g., no_smoking). Scores returned for each class correlate with the model's certainty the image meets our ground truth definition for the category. This page attempts to explain these ground truth definitions as clearly as possible.
The model makes classifications for each model head independently. In other words, if an image scores highly in multiple classes, the image meets our definitions for each class. Confidence scores from each model head are generated separately and are not correlated in and of themselves. It's easiest to think of this as asking multiple, narrower models (that may or may not overlap in scope) to each make a prediction on an image.
As a corollary, a high confidence score in negative classes (e.g., no_smoking) does not mean the image is clean in general. This is simply the logical opposite of the positive classification; the subject matter captured by the positive class in that model head is not present. For example, an image that scores 0.99 in no_smoking can still score highly in yes_pills, yes_illicit_injectables, or any other class trained to flag other subject matter. For this reason, we will describe these negative classes as a non-exhaustive list of subject matter that is not captured by the positive classes in that model head but that might be helpful to distinguish what is and is not flagged by that model head (e.g., borderline content, content captured by other classes).
For these classes, the model classifies animations, drawings, diagrams, paintings, or other artwork in the same way as photographs or photorealistic images.
The model head can be used to flag pictures of pills, including crushed pills and other drugs in power/crystal form. It does not flag other types of drugs such as injectables, IVs, and liquids.
yes_pills: the image shows pills or powder-like drugs that are clearly visible. This captures:
- Loose pills
- Pills visible within pill organizers, Rx containers, dosette boxes, pill cases or pill boxes
- Drugs in powder or crystalline form such as cocaine or MDMA
- Ecstacy and similar drugs in pressed pill format
- Crushed pills, half-crushed pills, or someone in the process of crushing pills
- Powders of any kind being snorted
no_pills: the image does not show clearly visible pills or powder drugs. For clarity, yes_pills does not capture:
- Bottles and containers intended to store pills with no actual pills visible.
- Drugs in liquid or injectable form
This model head can be used to flag images of injections and syringes. The model head defines two positive classes to differentiate between medical/prescription injectables and intravenous drugs used in recreational settings.
illicit_injectables: the image shows needles and syringes being handled or injected. The image also contains evidence of illicit, non-medical use, such as:
- Use of a makeshift tourniquet or belt
- Signs of being in a recreational or non-medical setting, such as outdoors, an alleyway, or a residence
- Non-medical paraphernalia such as spoons, lighters, or candles
- Presence of powders or crystals, alcohol, or other drugs
medical_injectables: the image shows needles and syringes being handled or injected. The image also contains evidence of medical use or a medical setting, such as:
- The person delivering the injection is wearing medical garb, such as scrubs, latex gloves, etc.
- The setting contains medical equipment such as a medical tourniquet, stethoscope, hospital bed, etc.
- The delivery device resembles a prescription injectable such as insulin shots, Epipens, etc.
- Veterinary injections
- Injectables/injections that are not obviously medical but the images has no indicators of illicit or recreational use
no_injectables: the image does not show injections, needles/syringes, or related drug paraphernalia. To be clear, non-injectable drugs such as pills, powders, inhalants, etc. are not covered by the other classes in this model head.
This model head can be used to flag images that show smoking, vaping, tobacco, marijuana, and related smoking paraphernalia.
yes_smoking: the image shows tobacco products, marijuana, or related delivery methods or paraphernalia, including:
- Cigarettes, cigars, marijuana joints, and similar paper-based smoking products
- Other tobacco products such as loose tobacco and dip
- Vapes, vape juice, and vape cartridges
- Marijuana plants and processed marijuana
- Grinders, alone or in use
- Bongs, pipes, hookahs, and similar smoking paraphernalia
no_smoking: the image does not show any of the above. Yes_smoking should not flag plants and herbs other than marijuana
This model head can be used to flag images depicting gambling, including casino games and slot machines or gambling on other games or events where a wager is clearly occurring. Pictures taken on a casino floor, or that include slot machines and similar games are always flagged as gambling. Games that can be played without gambling (e.g., card games) or games of chance are flagged as gambling ONLY IF money, chips, or similar tokens are present. These definitions also apply to animations, drawings, etc.
yes_gambling: the image depicts gambling in some format. This includes:
- Images taken on a casino floor
- Slot machines and pachinko
- Video poker, video blackjack machines or other gambling machines
- Online poker/blackjack sites or similar
- Street gambling such as jacks or dice games
- Any of the following if money, chips, or other tokens are clearly visible: cards and card games, dice, dominos, roulette or spin wheels, games with elements of luck such as backgammon or board games
- Evidence of betting on races or sporting events (paper receipts, screenshots of betting websites)
- Evidence of lottery participation (lotto tickets, scratchers, etc.) or lottery advertisements
- Raffles and bingo
no_gambling: the image does not depict gambling or evidence of gambling. For clarity, the following would not be flagged as gambling:
- Cards and card games, dice, dominos, board games, etc. with no money, chips or other evidence of wagering visible
- Photos from races or sporting events without paper receipts or proof of betting
- Entirely skill based games such as chess, even if money or tokens are present
- Arcade games and video games where money is not wagered
- Cards used in other contexts, such as magic tricks
- Trading cards, sports cards, etc.
This model head can be used to flag images of alcohol and alcohol use. Art, animations, and illustrations of alcohol or alcohol use are classified under a separate animated_alcohol class. Generally, images flagged as yes_alcohol will include obvious indicators of alcoholic beverages such as labeled cans or bottles, or contextual/setting information that indicates that beverages in the images are certainly alcoholic (e.g., photos taken at bars, parties, etc.)
yes_alcohol: the image is a photograph where alcoholic beverages are clearly visible or strongly implied. This includes:
- Cans and bottles that clearly contain(ed) beer, seltzer, hard cider, etc. indicated by bottle color, labels, logos etc.
- Glasses filled with what is clearly beer
- Wine bottles, indicated by shape, color, labels, etc.
- Filled wine glasses
- Liquor bottles, indicated by shape, labels, color of contents, etc.
- Filled shot glasses or pours of liquor
- Flasks that obviously contain alcohol (e.g., being poured)
- Cocktails, generally indicated by container, color, garnish, etc.
- Beverages that are not as certainly alcoholic but served in a bar or party setting
- Kegs and beer taps
- Wine taps and bagged wine
- Barrels that clearly contain wine or whiskey
- Photos of brewing, winemaking, distillation processes and/or equipment for the same clearly in use
- Obvious drinking games, even if alcohol containers, kegs, etc. are not visible
animated alcohol: an animated image, illustration, or art depicting the above alcoholic beverages or alcohol use.
no_alcohol: the image does not show alcoholic beverages or shows beverages that cannot be certainly identified as alcohol. The following is not flagged by yes_alcohol:
- Solo cups and other cups where the contents are not visible or identifiable as alcohol
- Bottles or cans in brown paper bags
- Flasks where the contents cannot be determined
- Cans without logos or labels identifying them as alcohol (may be flagged in a bar/party setting unless there is evidence it is non-alcohol
- Glassware typically used to serve alcohol (e.g., wine glasses, shot glasses, tumblers, etc.) that is not in use
- Sodas, sparkling water, etc. in clear glass bottles
- Other beverages that are clearly not alcohol such as juices, milk, etc.
- Smoothies and frozen drinks unless indicated to be alcoholic by contextual factors.
Updated 8 days ago