foamliu/mobile-image-matting

a lightweight image matting model

Deep Mobile Matting

This is a lightweight image matting model in PyTorch.

Features

1. MobileNetV2 as backbone.
2. DeepLabv3 heads.
3. Small model (size: 23.5MB, FLOPs: 11.39GB, total params: 7.62 millions)

Performance

• 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.

Models	SAD	MSE	Download
paper-stage0	59.6	0.019
paper-stage1	54.6	0.017
paper-stage3	50.4	0.014
my-stage0	127.4	0.068	Link

Dependencies

• Python 3.6.8
• PyTorch 1.3

Dataset

Adobe Deep Image Matting Dataset

Follow the instruction to contact author for the dataset.

MSCOCO

Go to MSCOCO to download:

• 2014 Train images

PASCAL VOC

Go to PASCAL VOC to download:

• VOC challenge 2008 training/validation data
• The test data for the VOC2008 challenge

Usage

Data Pre-processing

Extract training images:

$ python pre_process.py
# python data_gen.py

Train

$ python train.py

If you want to visualize during training, run in your terminal:

$ tensorboard --logdir runs

Experimental results

The Composition-1k testing dataset

1. Test:

$ python test.py

It prints out average SAD and MSE errors when finished.

Demo

Download pre-trained Deep Image Matting Link then run:

$ python demo.py

Image/Trimap	Output/GT	New BG/Compose

mobile-image-matting's Issues

Could you please provide the pre-trained model for demo?

Hi,
Thanks for you work!
I found that there is no pre-process.py file. And the pre-trained model does also not exist. As a result, I can not run the demo.
Could you please release your trained model weights file as well as other necessary files (if there are) for demo?
Thanks in advance.

Da

404 error on the Checkpoint link

Which is the correct link for the smaller BEST_checkpoint.tar file for mobile matting?

1. In the Demo section of ReadMe, it gives the following link for the pretrained model, but it gives a 404 error

https://github.com/foamliu/Deep-Mobile-Matting/releases/download/v1.0/BEST_checkpoint.tar

2. In the Performance section of ReadMe, it gives following link in the table against my-stage0

https://github.com/foamliu/Deep-Image-Matting-v2/releases/download/v1.0/BEST_checkpoint.tar

But the above points to the large checkpoint file from the main "Deep-Image-Matting-v2" project.

Trying to evaluate on this second link give following error

AttributeError: Can't get attribute 'DIMModel' on <module 'models' from '/content/gdrive/My Drive/MobileImageMatting/Mobile-Image-Matting/models/__init__.py'>

cannot find training_fg_names.txt

Bad training loss for "mobile net" backbone in deeplab

I trained the demo with "mobile net" backbone, but it could only reach to a alpha prediction loss of 0.1220 or so, which is far worse than "segnet".

Why should this happen? What is your loss? Thanks!

Create Google Colab notebook for better understanding

I am trying to implement this approach in my project. I doubt that do I need two input images one with a background with the subject and one is the background I need to replace with.

if you will make a Google Colab Notebook as an example it will be very helpful to implement.

The download link for pre-trained model isn't working anymore.

Error: Page not found

Integrating Mobile_Image_Matting into Android Project

Hi,
Thanks for you work!
its looking awesome output.
I want to integrate your demo into android project. Is it possible to integrate model into android Project? If it possible, then How can i integrate this model into android project?
Can you please give some suggestions?
Thanks in advance.

pre trained model is not found on a given link?

can you please reshare the model link on [email protected]

Link to the pretrained model

Hello,

The link to the pretrained model is broken. Could you please provide a new one?

run the demo.py failed

I was download the BEST_checkpoint.tar, and run the demo.py, but something wrong:

Traceback (most recent call last):
File "demo.py", line 69, in
checkpoint = torch.load(checkpoint)
File "/home/search/miniconda3/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py", line 426, in load
return _load(f, map_location, pickle_module, **pickle_load_args)
File "/home/search/miniconda3/envs/pytorch/lib/python3.7/site-packages/torch/serialization.py", line 613, in _load
result = unpickler.load()
AttributeError: Can't get attribute 'DIMModel' on <module 'models' from '/data1/search/Mobile-Image-Matting/models/init.py'>

and I can't found any "DIMModel" in the repo?