search

lxtGH / CAE

This is a PyTorch implementation of “Context AutoEncoder for Self-Supervised Representation Learning"
193 stars 22 forks source link

BUG #2

Closed ross-Hr closed 2 years ago

ross-Hr commented 2 years ago

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 commented 2 years ago

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

ross-Hr commented 2 years ago

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 commented 2 years ago

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 commented 2 years ago

. |-- 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 commented 2 years ago

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)