Simulated appearance VS. simulated association scenarios (IMAGE)
Caption
(a) Simulated appearance usually has an image-style discrepancy with the real-world appearance. For many appearance-centered tasks such as re-identification, such appearance domain gap compromises models that are trained on synthetic data and tested on real data. (b) In comparison, we show that synthetic data are as effective as real data in training association models. It suggests that association scenarios (e.g., trajectories and occlusions) have a small domain gap between the synthetic and the real.
Credit
Beijing Zhongke Journal Publising Co. Ltd.
Usage Restrictions
Credit must be given to the creator.
License
CC BY