This dataset consists of video recordings of finger-pointing actions. The videos capture the complete movement of the index finger pointing from outside the camera lens towards the printed English words on a book. The finger must remain on the book for at least 2 seconds. The light sources used during the collection include LED lights and incandescent lamps, and the light coverage includes strong, medium, and weak types. The camera captures both the left and right hands. The captured range of the fingers includes the entire hand and arm, the entire index finger, half of the index finger, and four types: the entire hand, the entire index finger, half of the index finger, and the entire hand plus the arm. The types of books also include those with doodles and clean books, with or without creases, with or without curling. The collection equipment covers 5 positions: the front, upper right, upper left, right side, and left side of the screen, and is shot in landscape mode. The main subjects of the collection are children and middle-aged people.
This dataset was collected using high-definition cameras in six supermarket scenarios to capture the self-checkout scenes. The collected data covered six self-checkout scenarios, with approximately 10 people being filmed in each scenario. The checkout equipment types included vertical self-checkout machines and handheld scanning checkout machines. Each person was filmed using these two types of equipment to capture the checkout actions. There were two main types of self-checkout actions: normal checkout and abnormal checkout (with unchecked items). The collection equipment captured from three angles, including the person's direct above, the left above (20 degrees or 45 degrees), and the right above (20 degrees or 45 degrees). Based on the collected videos, three types of annotations were made: target item recognition detection box annotation - handheld device, target item recognition detection box annotation - self-checkout device, and video segment start and end frame annotation.
The data collection equipment for this dataset is a 3D space scanning device. It was collected in Chinese supermarkets. Using a 3D space scanning camera device, 10 complete 3D scene data of medium and small-sized supermarkets were collected, and a 3D point cloud data model was generated for each supermarket.
This dataset consists of 7 types of portrait video data, including selfies, live broadcasts, movies, TV shows, cartoons, anime, and variety shows. The video frames are extracted and labeled with semantic segmentation for human faces. The annotations cover both real humans and non-real humans. The product library contains labeled images covering normal light, dim light, and backlight conditions. The figures in the images include close-up photos, half-body photos, and full-body photos, and the poses are diverse, including frontal, side, and back views. The number of people in selfies and live broadcasts is mainly single, while other categories may have multiple people types (2-3 people). The overall data is mainly single-person. The categories of facial semantic segmentation annotations for the figures in the images are approximately 23 types.
This dataset collects live attack video data of Chinese people. The collection scenarios cover indoor and outdoor environments and various light conditions. The shooting distance will be dynamically adjusted according to the actual shooting situation, mainly changing between full-body shots and head-and-shoulder shots. The types of attack methods collected include static attacks and dynamic attacks, totaling 7 categories. Static attack forms include photos, A4 color prints, and black-and-white prints. Dynamic attack forms include playing videos on iPads, Android phones, iPhones, and laptops. Each person will shoot 6 live videos and 22 attack videos in a normal and natural state, with each video lasting approximately 15 seconds. The collection tools are collected using a handheld method.
