Floc image dataset from a ten-month jar-test campaign at the Yoshimi Water Purification Plant (December 2018 - September 2019), with CNN train/test partition and analysis code

Floc images (300 x 300 px, 121 frames per jar-test video at 1 frame per second from the 4-6 min window) from jar tests of natural raw water sampled twice weekly at the Yoshimi Water Purification Plant (Ara River, Japan) from December 2018 to September 2019, with the video-level-exclusive train/test partition (342 training videos / 86 held-out test videos), per-image class labels defined by final supernatant turbidity, model predictions, weekly performance records, feed-water quality, the jar-test-grouped five-fold cross-validation fold assignment and pooled out-of-fold predictions, and the executed training/evaluation notebooks. This deposit supports the revision of manuscript WR115802 (Water Research, under review): "Seasonal evaluation of a convolutional neural network for coagulation assessment from floc images at operating water treatment plants." See README.md for full documentation, caveats, and licensing (data CC BY 4.0; code MIT).