Object recognition uses AI to detect objects in a photo and auto-tag those photos to make finding them easier.
Adding images to a system that utilizes object recognition can make finding images significantly easier. Adding meaningful tags to images, without having to do so manually, is an opportunity to greatly increase the value of an image collection.
When I’ve added images, such as the first example below, to DBGallery I am always easily able to find photos of Water, Shack and Outdoors, or Buildings afterward without any manual tagging. Recognition detects these scenes easily. It significantly adds to the overall value of my image collection. After all, what’s the use of a photo collection when photos can’t be found. You know you took photos of your company’s ice skating outing, but where are they?! Lost? If so, the photo collection’s value is just not there. Basic object recognition can make the collection so very much more valuable!
Objects detected below: Nature, Outdoors, Water, Building, Countryside, Hut, Rural, Shack, Land, Shoreline
Objects detected below: Building; Viaduct; Bridge; Nature; Outdoors; Architecture; Water
How it works in DBGallery
You can try it out in a couple of ways. In order of simplicity:
Option #1. Drop images onto our extremely simply Object Recognition demo page.
Option #2. Log into dbgallery.cloud (Username: DAdmin, PW: dbg), go to any folder there and upload images. Be sure to click the Upload button, highlighted in Red below, and choose the “Automatically recognize objects in the new images” checkbox.
The above covers the very basics. If you’re interested in going to a little deeper, at the risk of confusing how simple it usually is, read on.
But it’s not perfect
In the second example sample above, the one with the bridge, it didn’t add pick up tags such as River or Castle. Then again those were not the focus of the photo. The above example used Amazon’s Rekognition as the object recognition service. Google’s Cloud Vision or IBM’s Watson, Azure Vision, Clarifai, and other image recognition systems would have all returned some variation on objects that were recognized. The reality is that some manual tagging is still required to get images tagged to a refined level. Object recognition can go anywhere from ‘a good start’ to ‘all I need’, depending on the types of objects in your images and important getting tags exactly right is.
DBGallery is working on a feature that allows users to choose which words to ignore, such as never tagging images with “Human”, as returned in the above Ice Skating example. Unfortunately there is no means of interacting with object recognition that gives indicators such as “season” or “weather”, where the AI would be sure to look for such traits and return the appropriate tags. Maybe sometime in the near future it will.
Custom models can be trained for recognizing the tags you’re looking for. This is done by feeding images of specific objects to the recognition system, teaching it what specific objects looks like. A common example is feeding the AI with images of various types of clouds. Upload a batch of images, perhaps 100 - 250 (the more the better) of Cumulus clouds, then a batch of Cirrus, Stratocumulus, etc. and it will learn to recognize those cloud types appropriately later as you add various images that have those types of clouds in them. This can work very well when you have images specific to your industry, such as for recognizing certain types of cells, including indicating good and bad ones, in doing cancer or other disease research. Or having it recognizing specialty tools or antique and specialty objects.
See more on the Custom-trained AI object detection in DBGallery blog post.
Not yet a part of DBGallery, facial recognition works similar to custom models. You upload one or more images of a given face, indicating the name of that person, and from there on when photos of that person are submitted for recognition, their name will be returned as a tag.
Where from here