Manifold_Segmentation

Manifold segmentation is a method to capture the context information in the image through manifold regularization. The method with Resnet and MolibeNetv2 backbones for Pytorch.

Manifold_Segmentation's applications

As a regularization item, the model can be widely used in image segmentation and has been proven effective.

code

Install dependencies

python -m pip install -r requirements.txt

This code was testd with python 3.7

Train

Visualize training (Optional)

Start visdom sever for visualization. Please remove '--enable_vis' if visualization is not needed.

# Run visdom server on port 28333
visdom -port 28333

run the train code

python main.py --model 'model' --dataset 'dataset_name' --data_root 'path' --enable_vis --vis_port 28333 --gpu_id 0 --lr 0.01 --mr 0.001 --crop_size 'size' --batch_size 16 --download

Specify the model architecture with '--model ARCH_NAME'

DeepLabV3	DeepLabV3+	Head	Doubleattention
deeplabv3_resnet50	deeplabv3plus_resnet50	head_resnet50	doubleattention_resnet50
deeplabv3_resnet101	deeplabv3plus_resnet101	head_resnet101	doubleattention_resnet101
deeplabv3_mobilenet	deeplabv3plus_mobilenet	:---:	:---:

dataset download

cityscapes

Select parameters

--dataset cityscapes  -- download

voc

Select parameters

--dataset voc --year 2012_aug  -- download

Continue Traing

Run main.py with '--continue_training' to restore the state_dict of optimizer and scheduler from YOUR_CKPT.

--ckpt YOUR_CKPT --continue_training

Test

Results will be saved at ./results.

python main.py --model 'model' --enable_vis --vis_port 28333 --gpu_idch 0 --lr 0.01 --mr 0.001 --crop_size 'size' --batch_size 16  --ckpt 'model path,such as ./checkpoints/***.pth ' --test_only --save_val_results

Thanks to the Third Party Libs

DeepLabV3Plus-Pytorch Dual Attention Network
pytorch-segmentation-toolbox