Attention map visualization does not include alpha values that could make it more visually appealing

Currently the gifs being made for visualizing the attentions masks cover the image in black with scaled red rather than showing the page and mask together as, basically, a heat map of the attention. I have fixed this by adding alpha values, scaled on the output attentions maps, to make the areas of the image that are not "of attention" the colors that it should be instead of black (gray when the images are blended). I have also added arguments that can take user input for different color mappings and a scaling factor that determines how intensely the attention maps are shown. To do this, I needed to make changes within the dan/ocr/predict folder, add the requirement of matplotlib (for color map creation). The code can be found here - https://gitlab.teklia.com/atr/dan/-/tree/fix-attention-gif-transparency?ref_type=heads