Mitigation of Effects of Occlusion on Object Recognition with Deep Neural Networks through Low-Level Image Completion

Figure 5

Classification results on SORBO as a function of occlusion level and classification algorithm with the combined training set. The occlusion bins partition the testing samples as defined in Figure 3. Consistent with the analysis of Hoiem et al. [9], classification performance decreases with increasing occlusion and drops to a level marginally better than chance in high-occlusion conditions.