third_party/DOWNLOAD_DATA.md

## Download datasets

### VOC2007
To download [Pascal VOC 2007 train and validation data](http://host.robots.ox.ac.uk/pascal/VOC/voc2007/): 
```
cd datasets/VOC2007/
bash download_voc07.sh
```
The dataset should be organized as : 

```
./datasets/VOC2007/VOCdevkit
                    ├── VOC2007/
                        ├── JPEGImages
                        ├── Annotations
                        ...
```
### VOC2012

To download [Pascal VOC 2012 train and validation data](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/): 
```
cd datasets/VOC2012/
bash download_voc12.sh
```
The dataset should be organized as : 

```
./datasets/VOC2012/VOCdevkit
                    ├── VOC2012/
                        ├── JPEGImages
                        ├── Annotations
                        ...
```

### COCO
To download [COCO dataset](https://cocodataset.org/#home): 
```
cd datasets/COCO/
bash download_coco.sh
```
The dataset should be organized as : 
```
./datasets/COCO/
               ├── images/
               ├── annotations/
```

### CUB
To download [CUB 200-2011](http://www.vision.caltech.edu/visipedia/CUB-200-2011.html):
```
cd datasets/CUB/
bash download_cub.sh
```

The data should be organized as
```
./datasets/CUB/CUB_200_2011
                    ├── images/
                    ├── images.txt
                    ├── image_sizes.txt
                        ...
```
The `image_sizes.txt` file can be generated by `/dataset/CUB/compute_image_size.py` 


### ImageNet
The data should be organized as
```
./datasets/ImageNet/
                    ├── ILSVRC
                        ├── Annoations
                        ├── Data
                        ├── Detection
                        ├── ImageSets
                    ├── LOC_synset_mapping.txt
                    ├── LOC_val_solution.csv

                        ...
```

### ECSSD
To download [ECSSD](https://www.cse.cuhk.edu.hk/leojia/projects/hsaliency/dataset.html):
```
cd ../datasets/ECSSD
bash download_data.sh
```

The dataset should be organized as:

```
../datasets/ECSSD/
                  ├── img
                  ├── gt
```

### DUTS
To download [DUTS](http://saliencydetection.net/duts/#org3aad434):
```
cd ../datasets/DUTS_Test
bash download_data.sh
```

The dataset should be organized as:
```
../datasets/DUTS_Test/
                     ├── img
                     ├── gt
```


### DUT-OMRON
To downlaod [DUT_OMRON](http://saliencydetection.net/dut-omron/#org96c3bab):
```
cd ../datasets/DUT-OMRON
bash download_data.sh
```

The dataset should be organized as:
```
../datasets/DUT-OMRON/
                     ├── img
                     ├── gt
```


## Download DINO checkpoint

### DINO pretrained checkpoints

We initialize the model weights by using dino pretrained weights. To download the dino model, please launch the following commands:
```
wget https://dl.fbaipublicfiles.com/dino/dino_deitsmall16_pretrain/dino_deitsmall16_pretrain.pth # vit-s/16
wget https://dl.fbaipublicfiles.com/dino/dino_deitsmall8_pretrain/dino_deitsmall8_pretrain.pth # vit-s/8
wget https://dl.fbaipublicfiles.com/dino/dino_vitbase16_pretrain/dino_vitbase16_pretrain.pth # vit-b/16
```

### DINO finetuned linear classifiers on ImageNet

Since Dino offered finetuned linear classifiers on ImageNet, we can simply download them by following commands:
```
wget https://dl.fbaipublicfiles.com/dino/dino_deitsmall16_pretrain/dino_deitsmall16_linearweights.pth #vit-s/16
wget https://dl.fbaipublicfiles.com/dino/dino_deitsmall8_pretrain/dino_deitsmall8_linearweights.pth # vit-s/8
wget https://dl.fbaipublicfiles.com/dino/dino_vitbase16_pretrain/dino_vitbase16_linearweights.pth # vit-b/16
```