Open xiaobao886 opened 3 years ago
You do need background data https://grail.cs.washington.edu/projects/background-matting-v2/#/datasets
您确实需要背景数据 https://grail.cs.washington.edu/projects/background-matting-v2/#/datasets
Thank you for your reply Please ask dataset VideoMatte240K and PhotoMatte85,Whether the corresponding background image is not provided as in the sample set。So I can't run the training code train_base?
Please ask dataset VideoMatte240K and PhotoMatte85,Whether the corresponding background image is not provided as in the sample set。So I can't run the training code train_base?
Or do you want to combine the foreground and alpha images of the published dataset with the background image, and then put the combined image and background into the training?
Just download the dataset. Update data_path.py
, and the training script will composite images on the fly
Thanks for your reply
As you said, the images were made by composition, so how is this different from Adobe datasets?
The background description did not publish the training background picture, whether I still can not perform train_base and train_refine
You do need background data https://grail.cs.washington.edu/projects/background-matting-v2/#/datasets
Hi is this background datasets for train or test? I see no split fold for train and test? Thanks!
i have no idea how to organize the training data, do you solve this question and could you please tell me which one should i download and use for training ?
@Yiir-1 how did you use the checkpoints from stage 1 in stage 2. Can you provide full command to run stage 2?
I have a question. Background input is required in sample and paper, so do I need to input background during training? I see no background images in the publicly available data sets.
I want to run train_base and train_refine, The training results appear the following problems![IMG20210530135612_BURST000_COVER](https://user-images.githubusercontent.com/72678700/120094473-2741c480-c153-11eb-83ee-3871028d8e15.jpg)