Abundant Inverse Regression using Sufficient Reduction and its Applications
Temperature Prediction
Scene ID
Temperature prediction results on the Hot or Not dataset.
Temperature is predicted as T = T(lo) + T(hi), where T(lo) is the low-frequency
(e.g., seasonal) temperature, and T(hi) captures the day-to-day temperature variations.
Note that the weights change dynamically to reflect the characteristics of
each image in a sequence. Patches with high weights are more relevant to the
temperature prediction. Note that weights on occluded regions are generally
down-weighted.
