Accurate and comprehensive measurement of household livelihoods is critical for monitoring progress toward poverty alleviation and targeting social assistance programs for those who most need it. However, the high cost of traditional data collection has historically made comprehensive measurement a difficult task. This paper evaluates alternative satellite-based deep learning approaches using detailed household census extracts from four African countries to accelerate progress toward comprehensive, fine-scale, and dynamic measurement of asset wealth at scale. The results indicate that transformer architectures solve multiple open measurement problems, by providing the most accurate measurement of local-level variation in household asset wealth across countries and cities, as well as changes in household asset wealth over time. Experiments that artificially restrict data availability show the model's ability to achieve high performance with limited data. The proposed approach demonstrates the promise of combining satellite imagery, publicly available geo-features, and new deep learning architectures for hyperlocal and dynamic measurement of wealth in data-scarce environments.