This dataset supports the paper entitled "Run-time Performance and Power Optimization of a Parallel Disparity Estimation Algorithm on Many-Core Platforms" accepted for publication in the journal ACM Transactions on Embedded Computing Systems (TECS). Data Supporting Figures/Tables: Table I - Comparison of the accuracy of related stereo matching algorithms using standard image pairs from the Middlebury database. Figure 3 - Power and performance trade-offs for the possible range of cores and operating frequencies when executing the DE algorithm. Table II - Normalized power and performance trade-offs. Figure 6 - Plots of measured power (a) and performance (b) with training data highlighted in red, and run-time models for power (c) and performance (d) generated from training data. Figure 7 - Evaluation of how increasing the number of training samples for both core count and frequency reduces the power and performance modeling error. Figure 8 - Run-time optimization examples (a) without performance constraint and (b) with performance constraint. Figure 9 - Time series analysis of the RTM performing online adaptations of core number and frequency to optimize power and energy whilst meeting a target performance. Table IV - Comparison of the power and energy savings for the proposed approach torecently reported RTM approaches. Table V - Run-time manager training and adaptation/reconfiguration overheads.