This repository regroups data associated with the manuscript entitled "Deep in situ microscopy for real-time analysis of mammalian cell populations in bioreactors". In this work, an in situ microscope (ISM) based on pulsed transmitted light illumination via optical fiber was combined to artificial-intelligence to enable an online cell classification according to well-known cellular morphological features. The labelled database, corresponding to the repository, was used for the training phase of a convolutional neural network devoted to the localisation and classification of cells. The model made it feasible to predict real-time the evolution of different classes of animal cells, among which viable, necrotic and apoptotic cells. Considered as a breakthrough in the catalogue of process analytical tools, this optical system powered by artificial-intelligence is also of great interest for research. The database is made up of images taken by ISM in suspended cultures showing viability values ranging from 90% to 20%. The annotation of 4 000 images by two interchanging experimenters was done according to the identification rule. The annotation of the database was done with using LabellImg version 1.8.4. For images that contained an object that coul be labelled according to the predefined identification rule, a txt file indicates the class of the object and its localisation. The ISM working principle is based on a pulsed transmitted light microscope with an objective *40, NA 75, object field 0.16 * 0.22 mm2, water immersion (LOMO, St. Petersburg, Russia) coupled to a camera CCD-size 2/3’’, 1 392 * 1 040 pixels corresponding to 6.6 * 8.8 mm2, pixel size 6.45 * 6.45 µm2 (A102f, Basler, Ahrensburg, Germany). A 300 µm glass fiber is combined to POF-coupled luminescence diode (DieMOUNT, Wernigerode, Germany). The frequency of evaluated camera frames is 5 per second. The size of cell-micrographs is 140 * 140 pixels, corresponding to 22.5 * 22.5 mm2. Pulses for transmitted light illumination via the optical fiber are generated by the luminescence diode. The suspension is separated from the objective by a quartz-glass window positioned opposite of the fiber ending. A gap between fiber-ending and quartz-surface allows the suspension to flow through freely. Because of the spatial vicinity of the mixing impeller, it is estimated that the suspension has a time mean velocity close to the top velocity of the impeller, i.e., 0.1 m/s in the study. The cells drifting in front of the window can be imaged on a CCD without blurring thanks to the flash illumination of the luminescence diode. Cells appearing with sharply imaged details belong to the probe volume. Historically, counting via depth of field was first reported by Jean Perrin in 1909 in an experiment on the microscopic observation of an emulsion. Individual cell analysis can thus be considered to study the heterogeneity of a population of cells. Only one single momentary image is obtained from a set of cells which happens to be moving through the observed probe volume during the short illumination flash. After each illumination flash, the flowing suspension immediately shifts a new cell-sample into the probe volume. Subsequent flashes will always image entirely new cell-samples. Hence, it is impossible to follow the dynamics of the morphology of one individual cell with ISM, nor its descendants.