In the appendix, we present details of the selective search algorithm, the ImageNet linear evaluation results and the broader impact of this work

A.1 Implementation Details There are mainly three parameters in the selective search approach: scale, σ and min_size. The parameter scale controls the number and size of the produced segments, that higher scale means less but larger segments. The parameter σ is the diameter of the Gaussian kernel used for smoothing the image prior for segmentation. We use the default values of this approach: scale = 500, σ = 0.9 and min_size = 10. A.2 Visualization Figure 1 shows the proposals generated by the selective search approach, which shows reasonably good to cover the objects.