Detecting objects and text in an image

[deapdaru]

Use of a multi-task network is required to perform object detection and text recognition on images.

[raj2729]

Important elements to monitor while detecting objects and texts in images.

[robschu]

Output: Find out what we can see in an image. What kind of labels can be found. What kind of segmentation can be detected.

Create list of requirements towards an extraction model to be used to compare the models in the future.

[deapdaru]

Below are specific features and their significance that are expected as output from the model that the input image of proof is fed into:

• recognize various objects that are present in the image and create a bounding box around them. Significance: to identify different boundaries of objects whether they are in plain sight or overlapped on top of one another, for instance within a vending machine the model must recognize the product items on display.
• label the object within these bounding box from a pool of classes (eg. brand logo, can, bottle, vending machine, car display, etc.), for which a list needs to be deliberated from the image dataset. Significance: recognizes the particular item within each bounding box which can be used to check whether the promotional product item is on display.
• segmentation of the object as detected by the label within the bounding box. Significance: to show within the box out much area is taken up by the item or if it gets obscured by another item in front of it, which can be useful to find out performance of the promotion. (Possibly can also be used to resolve issue #7 later on.)
• text recognition needs to be performed on objects identified as "text". Significance: to read the brand name and further help in linking to the promotion.

Reference: