Deep Image Matting v2
Deep Image Matting paper implementation in PyTorch.
- "fc6" is dropped.
- Indices pooling.
"fc6" is clumpy, over 100 millions parameters, makes the model hard to converge. I guess it is the reason why the model (paper) has to be trained stagewisely.
- The Composition-1k testing dataset.
- Evaluate with whole image.
- SAD normalized by 1000.
- Input image is normalized with mean=[0.485, 0.456, 0.406] and std=[0.229, 0.224, 0.225].
- Both erode and dialte to generate trimap.
- Python 3.5.2
- PyTorch 1.1.0
Adobe Deep Image Matting Dataset
Follow the instruction to contact author for the dataset.
Go to MSCOCO to download:
Go to PASCAL VOC to download:
- VOC challenge 2008 training/validation data
- The test data for the VOC2008 challenge
Extract training images:
$ python pre_process.py
$ python train.py
If you want to visualize during training, run in your terminal:
$ tensorboard --logdir runs
The Composition-1k testing dataset
$ python test.py
It prints out average SAD and MSE errors when finished.
The alphamatting.com dataset
Download the evaluation datasets: Go to the Datasets page and download the evaluation datasets. Make sure you pick the low-resolution dataset.
Extract evaluation images:
$ python extract.py
$ python eval.py
Click to view whole images:
Download pre-trained Deep Image Matting Link then run:
$ python demo.py