The inspiration behind this bot came from everyday frustrations. I like learning about things. When someone tells me about something unfamiliar, I Google it. When I hear a song on the radio I like, I identify it through Soundhound or Shazam. But how about when I want to know the species of a plant, breed of a dog or model of a car? How am I supposed to describe it to Google?
That, is how WTFIT was born! Right now, it works well in identifying every day objects and animals. However as the community grows, the platform will become smarter and smarter, and be able to to identify even the toughest of images!
Report
@techno246 Concerning Google, I wonder how this classifier compares with Google Search by Image. Anyway, good job!
@mommi84 I've played around with Google search a lot, and it works very well when the image exists on Google images already. However, it doesn't seem to do as well with images you just take a photo, and the similar photos can sometimes be totally off. And thanks!
@vallieres@nagra__ you guys should check out Microsoft Cognitive Services, Google Cloud Vision and Cloudsight. They all have REST APIs for that purpose!
@techno246@nagra__ yes, I did have great success with Caffe from Berkley. Sadly we did not go through with the feature using it. We were hoping to crowdsource 1 year of photo and human taging as a baseline and have it fully automated afterwards.
Well, the recognition algorithm is not perfect. But I know a way you can improve it.
I uploaded an image of a potato soup in a plate (the file from Wikipedia had a name Romanian_potato_soup.jpg). At first I thought to rename the file because it would be too easy to know what was there even without looking at the picture. But the answer was too general - "A white plate". It means that the algorithm doesn't take file names into account, when they could be really useful.
@techno246 Very cool concept - I dig it. As mentioned by other commenters, it's pretty accurate in giving a very generalized identification (e.g., photo of a jade plant returned something like "leafy green potted plant"), which is cool, but I'm guessing most people know the generalized identification already -- getting more granular is key.
And while I do like the name, you should consider changing it to something a bit more kid-friendly, maybe just "WIT" (What Is That). Gotta love a good double-entendre!! Good luck.
@adammash Thanks for the feedback. Once we get experts on board the platform, we might be able to begin identifying specific traits of objects like species, etc. The community can patch up things. Out of interest, how willing would you be to contribute back to tagging images? For example if the bot hit you up an image, would you help identify it if you knew what it was? Something we are currently exploring.
Oh yes, I have thought about how to make it more kid friendly. Though it's a fun name and resonates with a lot of people, I wonder if I can keep it as a double entendre but have a kid friendly meaning of the acronym. WIT could work... Given W can still be expanded to it's original form, if needed :)
@techno246 I definitely think asking experts to contribute to tagging is a smart idea. It's how Google images tagged their images -- they had a "game" where people could flip through images and add any tags they think are relevant. I'm guessing they showed each image to multiple people and kept the tags that were suggested more than once. WIT! 😎
Report
Very fast, it did confuse my putty knife for a spatula but quite impressive.
Gift Insights
Gift Insights
Bagel
Gift Insights
Crypterval
RGB – #6: Rule Them All... Maybe Not
Gift Insights
RGB – #6: Rule Them All... Maybe Not
NotAnEasyChoice
Gift Insights
Gift Insights
Being
Gift Insights
Being
Gift Insights