BUG

[ross-Hr]

Hi, i just two suggestions when i run your CAE:
(i)
Your data dir is ' data_root = '/root/paddlejob/workspace/env_run//' '.
So there could be some unknown attacks.

(ii)
Your structure of 'downstream_task/detection/evaluation/object_detection' is awful, so that i run the code directly in my server and get an import BUG .

[SelfSup-MIM]

Thanks for your reminder! We have fix (i). Could you provide more details of issue (ii)?

[ross-Hr]

Thanks for your reminder! We have fix (i). Could you provide more details of issue (ii)?

In the directory : .../downstream_tasks/detection/evaluation/object_detection :

I create a.sh or b.sh in detection/evaluation/object_detection for training. It seems like :
....
python -m torch.distributed.launch --nproc_per_node=2
....

I also move the library detection/models into mmcv_custom/models.

[SelfSup-MIM]

There is no need to move models to mmcv_custom. We follow ibot to run object detection experiments.
Please enter downstream_tasks/detection/ and run your scripts in this folder.

[ross-Hr]

.
|-- init.py
|-- a.sh
|-- a.txt
|-- b.sh
|-- configs
| |-- base
| | |-- datasets
| | | -- coco_instance.py
| | |-- default_runtime.py
| | |-- models
| | | |-- cascade_mask_rcnn_r50_fpn.py
| | | |-- cascade_mask_rcnn_swin_fpn.py
| | | |-- cascade_mask_rcnn_vit_fpn.py
| | | |-- mask_rcnn_r50_fpn.py
| | | -- mask_rcnn_vit_fpn.py
| | -- schedules
| | -- schedule_1x.py
| -- mask_rcnn
| |-- vit_base_giou_4conv1f_coco_maskrcnn_1x_cae_sincos_init0.1_lr00003.py
| -- vit_large_giou_4conv1f_coco_maskrcnn_1x_cae_sincos_init0.1_lr00002_lrdr0.85_dp0.2.py
|-- mmcv_custom
| |-- init.py
| |-- pycache
| | |-- init.cpython-36.pyc
| | |-- checkpoint.cpython-36.pyc
| | |-- layer_decay_optimizer_constructor.cpython-36.pyc
| | |-- prepare_rpe.cpython-36.pyc
| | |-- register_backbone.cpython-36.pyc
| | -- utils.cpython-36.pyc
| |-- checkpoint.py
| |-- layer_decay_optimizer_constructor.py
| |-- loader.py
| |-- models
| | |-- init.py
| | |-- pycache
| | | |-- init.cpython-36.pyc
| | | |-- swin_transformer.cpython-36.pyc
| | | -- vision_transformer.cpython-36.pyc
| | |-- head.py
| | |-- swin_transformer.py
| | -- vision_transformer.py
| |-- prepare_rpe.py
| |-- register_backbone.py
| |-- runner
| | |-- init.py
| | |-- pycache
| | | |-- init.cpython-36.pyc
| | | |-- checkpoint.cpython-36.pyc
| | | -- epoch_based_runner.cpython-36.pyc
| | |-- checkpoint.py
| | -- epoch_based_runner.py
| -- utils.py
|-- output_dir
| |-- 20220630_151058.log
| |-- 20220630_151058.log.json
| |-- epoch_1.pth
| |-- epoch_2.pth
| |-- latest.pth -> epoch_2.pth
| -- vit_base_giou_4conv1f_coco_maskrcnn_1x_cae_sincos_init0.1_lr00003.py
|-- test.py
-- train.py

13 directories, 48 files

[ssocean]

add code below to the very first of the script could be more than enough~
import sys
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(file)))
sys.path.append(BASE_DIR)