PiCANet-Implementation

Pytorch Implementation of PiCANet: Learning Pixel-wise Contextual Attention for Saliency Detection

PPT(Korean)

https://www.slideshare.net/JaehoonYoo5/picanet-pytorch-implementation-korean

Top 10 Performance Test with F-score (beta-square = 0.3)

Step	Value
55000	0.827128
123000	0.837655
130000	0.825695
139000	0.823819
159000	0.824045
180000	0.833425
194000	0.829462
218000	0.823438
227000	0.823538
243000	0.825455

Execution Guideline

Requirements

Pillow==4.3.0
pytorch==0.4.1
tensorboardX==1.1
torchvision==0.2.1
numpy==1.14.2

My Environment

S/W
Windows 10
CUDA 9.0
cudnn 7.0
python 3.5
H/W
AMD Ryzen 1700
Nvidia gtx 1080ti
32GB RAM

Docker environment

docker run --name=torch -it -rm pytorch/pytorch
pip install tensorboardX
pip install datetime
git clone https://github.com/Ugness/PiCANet-Implementation

Docker image(Unstable)

https://hub.docker.com/r/wogns98/picanet/
based on pytorch/pytorch
codes in /workspace/PiCANet-Implementation
You can run code by add images and download models from google drive

You can run the file by following the descriptions in -h option.

python train.py -h

    usage: train.py [-h] [--load LOAD] [--dataset DATASET] [--cuda CUDA]
                [--batch_size BATCH_SIZE] [--epoch EPOCH] [-lr LEARNING_RATE]
                [--lr_decay LR_DECAY] [--decay_step DECAY_STEP]
optional arguments:
-h, --help            show this help message and exit
--load LOAD           Directory of pre-trained model, you can download at 
                    https://drive.google.com/file/d/109a0hLftRZ5at5hwpteRfO1
                    A6xLzf8Na/view?usp=sharing
                    None --> Do not use pre-trained model. 
                    Training will start from random initialized model
--dataset DATASET     Directory of your DUTS dataset "folder"
--cuda CUDA           'cuda' for cuda, 'cpu' for cpu, default = cuda
--batch_size BATCH_SIZE
                    batchsize, default = 1
--epoch EPOCH         # of epochs. default = 20
-lr LEARNING_RATE, --learning_rate LEARNING_RATE
                    learning_rate. default = 0.001
--lr_decay LR_DECAY   Learning rate decrease by lr_decay time per decay_step, default = 0.1
--decay_step DECAY_STEP
                    Learning rate decrease by lr_decay time per decay_step, default = 7000

python Image_Test.py -h

    usage: Image_Test.py [-h] [--model_dir MODEL_DIR] [-img IMAGE_DIR]
                         [--cuda CUDA] [--batch_size BATCH_SIZE]
optional arguments:
  -h, --help            show this help message and exit
  --model_dir MODEL_DIR
                        Directory of pre-trained model, you can download at
                        https://drive.google.com/drive/folders/1s4M-_SnCPMj_2rsMkSy3pLnLQcgRakAe?usp=sharing
  -img IMAGE_DIR, --image_dir IMAGE_DIR
                        Directory of your test_image ""folder""
  --cuda CUDA           'cuda' for cuda, 'cpu' for cpu, default = cuda
  --batch_size BATCH_SIZE
                        batchsize, default = 4

Detailed Guideline

Pretrained Model

You can download pre-trained models from https://drive.google.com/drive/folders/1s4M-_SnCPMj_2rsMkSy3pLnLQcgRakAe?usp=sharing

Dataset

I used DUTS dataset as Training dataset and Test dataset.
You can download dataset from http://saliencydetection.net/duts/#outline-container-orgab269ec.

• Caution: You should check the dataset's Image and GT are matched or not. (ex. # of images, name, ...)
• You can match the file names and automatically remove un-matched datas by using DUTSdataset.arrange(self) method

Execution Example

Assume you train the model with

• current dir: Pytorch/
• Dataset dir: Pytorch/DUTS-TE
• Pretrained model dir: Pytorch/models/state_dict/07261950/10epo_1000000step.ckpt
• Goal Epoch : 100

python train.py --load models/state_dict/07261950/10epo_1000000step.ckpt --dataset DUTS-TE --epoch 100

Assume you test the model with

• current dir: Pytorch/
• Testset dir: Pytorch/test
• Pretrained model dir: Pytorch/models/state_dict/07261950/10epo_1000000step.ckpt
• CPU mode

python Image_test.py --model_dir models/state_dict/07261950/10epo_1000000step.ckpt --img test --cuda cpu

Directory & Name Format of .ckpt files

"models/state_dict//<#epo_#step>.ckpt"

• The step is accumulated step from epoch 0.
• If you want to change the format of pre-trained model, you should change the code in train.py line 61-65

  start_iter = int(load.split('epo_')[1].strip('step.ckpt')) + 1
  start_epo = int(load.split('/')[3].split('epo')[0])
  now = datetime.datetime.strptime(load.split('/')[2], '%m%d%H%M')
  

Test with Custom_Images

• When you run Image_Test.py with your own Images, the images will saved in tensorboard log file.

• Log files will saved in log/Image_test

• You can see the images by execute

  tensorboard --logdir=./log/Image_test
  

  and browse 127.0.0.1:6006 with your browser(ex. Chrome, IE, etc)