pgt4861
pgt4861
@dajes thank you for your nice work! :D
Hi, we use 512x512 images when training the model. But you can put any size of image you want (as long as your gpu memory allows)
Hi, Thank you for your interest. We used all coco2014 training images as backgrounds and randomly sampled them when training. Following the default training configurations in the code is recommended.
Hi, @WLucky I'm sorry that I did not describe the detailed experimental environment about GPUs. In our experiments, we used RTX Titan or P40 which has 24G gpu memory. Thank...
hi, you can find and download the pretrained model on the Readmd.md thank you!
Hi @bfan1256, As you suggested, i think it can be processed frame by frame. In that case, we need a trimap per frame. If the trimap is given well, i...
hi, thank you for your interest on our work. I think the best way to draw trimap for separating the object (a cup) from the table is like below. In...
Hi, I'm sorry for the late reply. Some parts of our codes and training settings are originated from MG-Matting (as I wrote down on the readme.md). Specifically, the config of...
Hi @gohar-malik Thank you for you interest on our work. I think there are some ways to improve the results on real-world applications. First, there are some domain gaps between...
Hi @lurchycc, I’m sorry for late response You can test your image by modifying the inferece.py. you need to prepare trimap corresponding to the image.