Image Inpainting with Learnable Bidirectional Attention Maps

Image Inpainting by End-to-End Cascaded Refinement with Mask Awareness

“…PConv [48] proposed partial convolution where the convolution is masked and re-normalized to be based on only valid pixels. [49] improved the partial convolutional network by a learnable attention map for feature re-normalization and mask updating. [15] proposed a gated convolution (GConv) which generalized partial convolution by providing a learnable dynamic feature selection mechanism for feature maps at each level.…”

“…Since the source code of Partial convolution [48] (Pconv) is not available, we implement it with the experimental settings in the paper. Bidirectional Attentional Maps [49] (LBAM) is tested with the source code and the pertrained models provided by its authors. Performances of PEN-Net [9] and StructureFlow [13] are also evaluated using their official released codes and models.…”

“…Comparison experiments are conducted with irregular masks and regular masks respectively. Due to the measurements are sensitive to the size of image, we test all the methods with the same image size 256×256 following the common practice [10], [11], [49]. Table I shows the evaluation results.…”