rhfeiyang/i2pnet_segformer

using segformer as backbone for "Intention-aware Information Propagation Network for Interactive Segmentation"

Intention-aware Feature Propagation Network for Interactive Segmentation (segformer backbone)

Here is the code for training and evaluation. Before that, you need to prepare data by downloading from the following link:

Dataset	Description	Download Link
MS COCO	118k images with 1.2M instances (train)	official site
COCO+LVIS*	99k images with 1.5M instances (train)	original LVIS images + our combined annotations
SBD	8498 images with 20172 instances for (train) 2857 images with 6671 instances for (test)	official site
Grab Cut	50 images with one object each (test)	GrabCut.zip (11 MB)
Berkeley	96 images with 100 instances (test)	Berkeley.zip (7 MB)
DAVIS	345 images with one object each (test)	DAVIS.zip (43 MB)
Pascal VOC	1449 images with 3417 instances (validation)	official site
COCO_MVal	800 images with 800 instances (test)	COCO_MVal.zip (127 MB)

After downloading the data, please change the paths to the datasets in config.yml.

Training

1. Train coarse level network

python train_coarse.py models/iter_mask/coarse_segformer_cocolvis_itermask_3p.py --workers 4 --exp-name segformer_coarse --gpus 0

2. Train fine level network

python train_fine.py models/iter_mask/fine_segformer_cocolvis_itermask_3p.py --workers 4 --int-model segformer_corase --exp-name segformer_fine --gpus 0

Evaluation

python scripts/evaluate_model.py --intention ./experiments/iter_mask/resnet34/hrnet_coarse/checkpoints/last_checkpoint.pth --segmentation ./experiments/iter_mask/resnet34/hrnet_fine/checkpoints/last_checkpoint.pth --datasets GrabCut,Berkeley

Due to the double blind and file size limitation, we can't provide the pretrained checkpoint now. We will provide the pretrained checkpoint later.

Performance