The dataset contains three different folders:
- ATEL1Frames: It contains the frames of a video captured using a surveillance camera in the natural environment using multiple subjects
- AT_Face_Coordinates: It contains the face coordinates of the subjects at a frame level. Frame no. is the index of the face location in the coordinate files.
- HR Gallery: It contains the high-resolution gallery images of each subject who participated in the data collection. 4 images of each participant are acquired for gallery-based face matching.
Each directory contains a test, Val, and train folders.
In both training and validation sets: the original directory refers to the clean images and pt_xxxxx refers to the turbulence-affected images. In the test set original and two different sets of turbulence images (namely XY and XY directories) are provided. The participants need to submit a score matrix corresponding to each set along with a max 2-page write-up describing their approach and performance on the training/Val set.
Instructions:
- The participants need to use only the training set for developing the network. Use of test sets is strictly prohibited.
- Avoid any form of cheating including manual annotations on the test. It is advised to develop ethical solutions.
- Participants are free to use any face cropping algorithm if they want to.
- Matching will only be performed using the HR gallery images provided.
- Participants need to submit the score matrix on the test set. For instance, if the test contains images of "N" subjects and contains M frames, then the score matrix will have the scores: (MxK)*(Nx4). Where 4 represents the number of gallery images of each N subject and k represents the number of subjects present in one frame.
- The evaluation will be performed both at the frame and video level. The score of videos will be calculated as the average scores of all the frames in that video.
- To obtain the dataset, participants need to sign the license agreement of the DroneSURF dataset (http://iab-rubric.org/index.php/dronesurf). Once the agreement is approved we will provide the link to the dataset of this challenge.
Data Link: Coming Soon.
Deadline: Score matrix submission: 05/15/2022