The four folders (d1, d2, d3, d4) contain the raw data of the four validations (d1, d2, d3 increasing the number of imeges; d4 using data augmentation). -- Each folder contains couples of files: one is named after the ID of the individual (for example, SA38) and the name of the experiment (EXP_4); the other has the same name + "hom". The "hom" csvs are the manual analisis of the video, the others are the output of the neural network. -- The "hom" files contain two columns: in the first there are the seconds of the video; the second column tell us if from the corresponding second on the fish is present (1) or absent (0). The output files have various columns: ID (ID of the fish), sec (number of the frame analized), X and Y (coordinates of the fish in the image), % (detected object ptobability or score), seconds (seconds of the video). -- The file "datos validacion TFM" contains the results of the analisis performed on raw data: Video (name of the video), Test (type of the experiment), Threshold (probability threshold used), False1 (percentage of wrong prediction of the portion of video where the fish is present), True1 (percentage of correct prediction of the portion of video where the fish is present), False0 (percentage of wrong prediction of the portion of video where the fish is absent), True0 (percentage of correct prediction of the portion of video where the fish is absent), False (percentage of total wrong predictions), True (percentage of total correct predictions), Validation (number of validation, from 1 to 4), Presence (percentage of presence of the fish in the video), Type (activity (normal) or exploration (juguete) video), Number of images (used for training), Augmentation (if we have used data augmentation or not). -- For the positioning validation, we randomply sampled from raw data "moments" of the video (raws) to extract the coordinates (X and Y) and to compare them with the corresponding frames of the video (not availble) as ground truth.

Peer reviewed