Multi-label Food Images Annotation Part 2

Multi-label Food Images Annotation The task consists in assigning a set of labels to food images. We are looking for images of “combo plates” (i.e. images that contain more than one food) as well as each of the dishes/ingredients appearing in them. Thus, we need you to assign multiple labels to a single image. At the same time, we also need to discard any image that only contains a type of food/dish or that the different foods/dishes appearing are mixed (i.e. which is NOT a combo plate). Check the attached image to see an example. CRITERIA FOR COMBO PLATE IDENTIFICATION The basic criteria for considering that an image contains a “Combo Plate” are the following: At least 2 different types of food appear in the same image/plate. All the types of food appearing are clearly separated (i.e. they are not mixed). The food shown does not have a specific well-known name and thus is not considered a typical dish for a certain region of the world (e.g. cuban-style rice, which usually contains egg, rice and banana clearly separated similar to a Combo Plate). On the image appears a standard serving of food for a single person (either on a single plate or on multiple plates). If multiple plates appear and the serving is not common for a single person then it must be labeled as “none”. INSTRUCTIONS FOR LABELING The following information will be given to the freelancer: 1) Url to the image labeling website. 2) Username and password for accessing to the website. 3) Set of jobs (groups of images) that need to be labeled. The jobs will contain around 1000 images each. 4) Link to a FAQ document that must be read to solve some basic doubts. The groups of images assigned to the Freelancer can contain any of food from a predefined set of dishes/ingredients (see sample image provided). This set, divided in 4 groups, will be available for selection during the annotation process. The freelancer will have to: 1) Access the image labeling website. 2) Investigate the visual appearance of all the dishes/ingredients that appear in the predefined set next to the displayed image. 3) Decide if the image appearing on screen is a “combo plate”. If it is a combo plate go to step 4a. If it is NOT a combo plate go to step 4b. 4a) Is a combo plate: select the top label “none” and go to step 5. 4b) Is NOT a combo plate: Assign as many labels to each of the pictures appearing on the screen by clicking on them with the mouse. The labels assigned to the shown picture is displayed in green. If a label is not selected for the current dish it is displayed in orange. If none of the labels apply to the current image, then the label “none” must be selected. 5) Click on the top right/left arrows to view the previous/next image and label them all. 6) Repeat points 3 to 5 until all images from all the assigned jobs have been labeled. If any of the images was labeled wrongly you can go back to double-check and correct any of the images’ labels. QUALITY ASSURANCE AND PAYMENT After finishing the image labeling process for the assigned jobs, the employer will check the quality of the job. If the quality of the labels is not enough, then the freelancer will be required to check the labels assigned. The payment will be received only after the quality assurance process. Each of the assigned jobs will contain a variable number of images (1000 images on average). The payment will depend on the total number of images belonging to the job that was assigned to the freelancer. The indicated project budget corresponds to the average payment received for labeling a complete job. If the freelancer labels the images of more than one job he/she will receive the amount corresponding to the total images labeled.