We present a catalog of visual like H-band morphologies of ∼ 50.000 galaxies (H f 160w < 24.5) in the 5 CANDELS fields (GOODS-N, GOODS-S, UDS, EGS and COSMOS). Morphologies are estimated with Convolutional Neural Networks (ConvNets). The median redshift of the sample is < z >∼ 1.25. The algorithm is trained on GOODS-S for which visual classifications are publicly available and then applied to the other 4 fields. Following the CANDELS main morphology classification scheme, our model retrieves the probabilities for each galaxy of having a spheroid, a disk, presenting an irregularity, being compact or point source and being unclassifiable. ConvNets are able to predict the fractions of votes given a galaxy image with zero bias and ∼ 10% scatter. The fraction of miss-classifications is less than 1%. Our classification scheme represents a major improvement with respect to CAS (Concentration-Asymmetry-Smoothness)-based methods, which hit a 20 − 30% contamination limit at high z. The catalog is released with the present paper via the Rainbow database

We quantify the morphological evolution of z ∼ 0 massive galaxies (M * /M ⊙ ∼ 10 11.2±0.3 ) from z ∼ 3 in the 5 CANDELS fields. The progenitors are selected using abundance matching techniques to account for the mass growth. The morphologies of massive galaxies strongly evolve from z ∼ 3. At z < 1, the population well matches the massive end of the Hubble sequence, with 30% of pure spheroids, 50% of galaxies with equally dominant disk and bulge components and 20% of disks. At z ∼ 2 − 3 however, there is a majority of irregular systems (∼ 60 − 70%) with still 30% of pure spheroids.We then analyze the stellar populations, SFRs, gas fractions and structural properties for the different morphologies independently. Our results suggest two distinct channels for the growth of bulges in massive galaxies.Around ∼ 30 − 40% were already bulges at z ∼ 2.5, with low average SFRs and gas-fractions (10 − 15%), high Sersic indices (n > 3 − 4) and small effective radii (R e ∼ 1 kpc) pointing towards an even earlier formation through gas-rich mergers or violent disk instabilities. Between z ∼ 2.5 and z ∼ 0, they rapidly increase their size by a factor of ∼ 4 − 5, become all passive and slightly increase their Sersic indices (n ∼ 5) but their global morphology remains unaltered. The structural evolution is independent of the gas fractions, suggesting that it is driven by ex-situ events.The remaining 60% experience a gradual morphological transformation, from clumpy disks to more regular bulge+disks systems, essentially happening at z > 1. It results in the growth of a significant bulge component (n ∼ 3) for 2/3 of the systems possibly through the migration of clumps while the remaining 1/3 keeps a rather small bulge (n ∼ 1.5 − 2). The transition phase between disturbed and relaxed systems and the emergence of the bulge is correlated with a decrease of the star formation activity and the gas fractions suggesting a morphological quenching process as a plausible mechanism for the formation of these bulges (although the eventual impact of major mergers and a growing black hole in the bulge should also be considered). The growth of the effective radii scales roughly with H(z) −1 and it is therefore consistent with the expected growth of disks in galaxy haloes.