Abstract: Concerning the problem that uniform photon tracing becomes inefficient when rendering scenes with difficult lighting condition, a visual importance based adaptive photon tracing approach is proposed. This approach firstly computes visual importance and constructs a visual importance map. Then, a novel importance function based on visual importance and photon path visibility is devised. Second, a hybrid method based on adaptive Markov chain Monte Carlo and replica exchange is proposed to sample the importance function and generate new photon paths. Finally, a new strategy for choosing sampling distributions is designed. Replica exchange probability is computed in advance and the sampling distribution is estimated according to the computed probability. Experimental results show that this approach can efficiently render the scenes with difficult lighting condition, while producing high-quality result.