You signed in with another tab or window.Reload to refresh your session.You signed out in another tab or window.Reload to refresh your session.You switched accounts on another tab or window.Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+9-4Lines changed: 9 additions & 4 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -70,7 +70,12 @@ We not only show the selected extremely hard samples in the test sets but also s
70
70
71
71
2. **Datasets preparation**
72
72
73
-
Download all the single train/test datasets from my [google-drive](https://drive.google.com/drive/folders/1jRJKv56QSa3gOp4w_64tVmzNMT_te-Kv?usp=sharing), or directly download the `datasets.zip` in the folder for all the data you need as following structure shows (COCO-SEG is too big, so you can download it separately). The file directory structure on my machine is as follows:
73
+
Download necessary datasets:
74
+
from my google-drive: [DUTS_class](https://drive.google.com/file/d/1SKaxMtIaLJk2CRdSbf-S0m6vMag1grmd/view?usp=drive_link), [COCO-9k](https://drive.google.com/file/d/1r6tRcSlvH8bXhaZD2VtGmHDxsXFl1v4z/view?usp=drive_link), [COCO-SEG](https://drive.google.com/file/d/1hkn2wP3uArctbst11XP4iKOCTa3tud5-/view?usp=drive_link), and [CoSOD_testsets](https://drive.google.com/file/d/1pTjxK4gu5kfVeR4Fdc1shZgk47FvybCe/view?usp=drive_link), or
75
+
from my BaiduDisk: [DUTS_class](https://pan.baidu.com/s/1xNUaar-bzS3apJpHQED9dg?pwd=PSWD), [COCO-9k](https://pan.baidu.com/s/1AEH593Sq1XGZHhgoT4fhfg?pwd=PSWD), [COCO-SEG](https://pan.baidu.com/s/1_1VmtOHBxDKp9qq55AARXA?pwd=PSWD), and [CoSOD_testsets](https://pan.baidu.com/s/136TGYw_dh7KtVAHw6Kgknw?pwd=PSWD).
76
+
The `CoSOD_testsets` contains CoCA, CoSOD3k and CoSal2015.
77
+
78
+
The file directory structure on my machine is as follows:
74
79
75
80
```
76
81
+-- datasets
@@ -95,19 +100,19 @@ We not only show the selected extremely hard samples in the test sets but also s
95
100
...
96
101
```
97
102
98
-
3. **Update the paths**
103
+
4. **Update the paths**
99
104
100
105
Replace all `/root/datasets/sod/GCoNet_plus` and `/root/codes/sod/GCoNet_plus` in this project to `/YOUR_PATH/datasets/sod/GCoNet_plus` and `/YOUR_PATH/codes/sod/GCoNet_plus`, respectively.
101
106
102
-
4. **Training + Test + Evaluate + Select the Best**
107
+
5. **Training + Test + Evaluate + Select the Best**
103
108
104
109
`./gco.sh`
105
110
106
111
If you can apply more GPUs on the DGX cluster, you can `./sub_by_id.sh` to submit multiple times for more stable results.
107
112
108
113
If you have the OOM problem, plz decrease `batch_size` in `config.py`.
109
114
110
-
5. **Adapt the settings of modules in config.py**
115
+
6. **Adapt the settings of modules in config.py**
111
116
112
117
You can change the weights of losses, try various *backbones* or use different *data augmentation* strategies. There are also some modules coded but not used in this work, like *adversarial training*, the *refiner* in [BASNet](https://openaccess.thecvf.com/content_CVPR_2019/papers/Qin_BASNet_Boundary-Aware_Salient_Object_Detection_CVPR_2019_paper.pdf), weighted *multiple output and supervision* used in [GCoNet](https://openaccess.thecvf.com/content/CVPR2021/papers/Fan_Group_Collaborative_Learning_for_Co-Salient_Object_Detection_CVPR_2021_paper.pdf) and [GICD](https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123570443.pdf), etc.