Google Image Labeler is a feature, in the form of a game, a Google Image that allows users to label random images to help improve the quality of Google image search results. It's online from 2006 to 2011 and relaunched in 2016.
Video Google Image Labeler
Histori
Luis von Ahn developed ESP Game, a game where two people are simultaneously given a picture, with no way to communicate, other than knowing the label that fits for each picture or pass signal. The ESP Game has been licensed by Google in the form of Google Image Labeler and launched this service, as beta on August 31, 2006.
Players pay attention to subtle changes in the game over time. In the early months, until around November 2006, players can see their individual guesses during play by moving images. When "congenital abuse" begins (see below) players can see if their partner is using those terms while the game is in progress. The game was changed so that only in the end of the game could a player click on "see partner's guess" and learn what he typed. "Congenita's abuse" was eventually stopped by a change in game structure in February 2007. During the first few months of 2007, regular players grew to recognize a group of images that marked a "robot" partner, always with the same label in the same order. This seems to have changed around March 13, 2007. Suddenly most of the pictures seen are brand new, and older images come with vast banned lists.
As of May 2007, there are substantial and substantial changes made to the game. Instead of 90 seconds, the player has 2 minutes. Instead of 100 points per image, the score varies to appreciate higher specificity. "Man" may get 50 points while "Bill Gates" may get 140 points. On 7 August 2007, another change was made: instead of just showing the point value of each match as the match took place, the value of each game was shown next to the matching word at the end of the game. This makes it easier to see the exact value of specific versus general labeling. Further changes were observed on October 15, 2007. The new version was enacted and then appears to have been withdrawn. In the new version, the player only sees the image he or she is labeling, while in the old version the image is collected at the bottom of the screen while the game is playing. Other changes are subtle; for example, the score is in the green letter in the new and red versions of the old one. The most significant change is that the clock freezes during the image changes, and the time is basically subtracted from the game's two minutes. The change seems to come into full effect on October 18, 2007.
In September 2011, Google announced it would stop a number of its products, including Google Image Labeler, Aardvark, Desktop, Fast Flip, and Google Pack. The game ends on September 16, 2011, to the discontent of many users. The game idea survives as an art annotation game on ARTigo.
Google launches Image Labeler in 2016. New services are less like games. Users choose a category and they show a series of images. They browse through each image and declare whether it is properly categorized.
Maps Google Image Labeler
Rules
Users are randomly paired with partners who are online and use the feature. Users can become registered players who collect a total score of all games played, or guests who play for one game at a time. Note that players from all over the world are allowed to play, and American and English English will be encountered (eg, soccer vs. football). When the number of uneven players is online, a player will play against a previously recorded set of words.
In the latest version of the game, the rules run as follows: During a two-minute period, the user and his partner show the same set of images and are asked to label as many as possible to describe each image. they see. When a user's label matches a partner label, both will earn points and move on to the next image until the time runs out. It's possible to forward the picture but both users must agree to do this. Scores vary from 50 to 150 depending on the specificity of the answer. A score of 150 is rare, but 140 points are given for a particular name or word, even if the word is spelled in the image. Terms with low specificity like "tree" or "human" only earn 50 points. There is no filtering for the truth, so if both players type "Jupiter" for a Saturn image, they may get 140 points.
Labels that have been previously approved by the user will be displayed on the "restricted" list and can not be used in that round. Some players think that the game is staggered by the image display, and that sometimes it takes the first words typed by one player to form a "forbidden" list for other players. In other words, forbidden words may be unilateral. This will explain the somewhat frequent situation when it seems the couple can not think of words like "car", "bird", or "girl". Very rarely, by the end of the game it will become obvious that one image is different for two players. One of the speculations is that this is just a mistake, while the other is that it is a test to see how fast people will pass when their descriptions do not match. It may also be a mechanism applied to view fraudsters, if the words for different images are similar. Sometimes, one user's computer will fail to load a new image or continue displaying the previous displayed image. This time likes also called to "pass" along on the part of both players.
End of game
After 120 seconds is over, the game ends. Users can see partner names for the first time, their scores for games (both credited) and their individual cumulative scores to date. This is compared to daily high scores for teams and individual high scores of all time. Google betting on the competitiveness of users to collect high scores to enlarge the number of image ratings.
The game's final screen also shows all the images presented during the game along with a list of labels being guessed by other players. The pictures themselves are then linked to the website where the images were found and can be explored for more information and to satisfy curiosity.
Benefits for Google
The game was not designed just for fun; it's also a way for Google to ensure that its keywords are matched to correct the image. Any matching words should help Google to build an accurate database used when using Google Image Search.
Without human image tagging, the Google Images search in the past depends on the image file name. For example, a photograph labeled "Bill Gates Portrait" may have associated "Bill Gates" as a possible search term. Google Image Labelers rely on humans that mark the meaning or content of the image, rather than the context of seeing where the image is used. By storing more information about images, Google stores more opportunities to find images in response to a user's search.
Issues and issues
Additional rules
- Some users complain that the rules are difficult to decipher. There is no such place mentioned, for example, that the player must press Return after typing the label.
- Beginners often make the mistake of typing some terms into the first box, not realizing that they are then considered, together, as phrases.
- The "pass" option is also not explained; although that means that players do not want to judge a word, some players think that this button should be pressed after guessing. One of these errors can easily produce a zero score.
Although these rules are not explained on the playback screen, some of them are actually explained in the Help page.
Other issues include:
- Images that fail to load, or load very slowly, use time from hours.
- Images should not be scaled down, then scaled to cause excessive pixilation and blur.
- Experienced users typed the letter "x" to avoid just skipping (and printing zero) when this happened. Note that "x" (or other words) will not work for two consecutive images so other terms ("empty" or "none") can also be used, as sometimes many images are not loaded.
- Only six labels appear on screen during a round; if more are added, some are scrolled from above. Observations show that all these labels are counted, and with a quick partner one can see nine or ten words for a single image. Additionally, after depressing "Passers" users will no longer see their partner labels count. If the user keeps typing, he can get a match on the words his partner has typed even when the user's word has passed. If one's partner types a matching word after the player has graduated, it will not count.
- Scores are usually low when some of the images presented have a number of "forbidden" tags. In some cases, the "disabled" tag is broad enough or complete, making it difficult for both partners to create a new, matching tag.
Abuse
Less than a month after launch, the game began to be misused. It seems Google gets spammed with words from the following list: abrasive, accretion, bequeath, carcinoma, congenita, diphosphonate, entrepreneurship, patience and googley . Because players can see their partner's response at the end of each round, they learn that other players use these words. Some then put these words into their answers for entertainment and score improvement. On or before February 7, 2007, Google changed the game to reduce abuse. The words in the list above are filtered. Also, certain images that have been the triggers for random words come immediately "Your Partner Wants to Pass." In a game revision March 13, 2007, "trigger images" were temporarily removed.
Criticism
A case study of Google Image Labelers looks at how assumptions about game play and game mechanisms lead to distrust of its players, some of which describe it as exploitative and deceptive. Users are not allowed to label images in any way other than points but they can learn new English words slowly.
References
External links
- Discussion and Tips from Players, commented on May 4 changes
- New Google Images Laber Site
Source of the article : Wikipedia
0 Comments