damo-cv / transfgu Goto Github PK

..
INFO - train_pascalvoc - train epoch: [0][ 457/ 457] Time: 1.62 (1.73) Loss(cat/unc/emb/all): 1.7187/0.8079/0.9578/2.9189 (1.9438/0.8700/0.9989/3.2037)
INFO - train_pascalvoc - * train epoch: [0] loss(cat/unc/emb/all): 1.9438/0.8700/0.9989/3.2037
ERROR - pascalvoc - Failed after 0:13:26!
Usage:
pascalvoc_trainval.py [(with UPDATE...)] [options]
pascalvoc_trainval.py help [COMMAND]
pascalvoc_trainval.py (-h | --help)
pascalvoc_trainval.py COMMAND [(with UPDATE...)] [options]
sacred.utils.MissingConfigError: eval_pascalvoc is missing value(s): ['exp_ckpt_dir']

Thanks for proposing such an interesting approach! I want to know how long will TransFGU take to train on COCO-S-27* with 4 × Tesla V100 32G GPUs?

Hi, thanks for the interesting work!

Could you please check/update the download links of Cityscapes Dataset at this project page? Many thanks!

Best,
Jun-Pu Zhang

I'm very sorry to bother you, but I can't solve this problem. I hope I can get your help.
When I run coco.py,ERROR:root:No token file found. Also make sure that a [prod] section with a 'token = value' assignment exists.
I used python=3.8 torch==1.10.1
I'm really looking forward to hearing from you, it's very important to me

Thanks for the innovation work.
We meet some problem when installing and train/val, e.g. scared problem, and conflict when installing kornia (0.5.8 seems OK).
Could you please provide the specific version in requirement.

I successfully reproduced the results on cocostuff27:
INFO - eval_coco - ACC - All: 44.5856
INFO - eval_coco - mIOU - All: 16.4780

how could I obtain the resutls on cocostuff27 with "curated" samples?

Hi,

Great work. I am interested to use this method on my custom dataset. Is it possible to train the model on custom dataset?

ERROR:root:No token file found. Also make sure that a [prod] section with a 'token = value' assignment exists.
I have a question for you，The error is to obtain token by CUDA_VISIBLE_DEVICES=0 python pascalvoc_crop.py，The generated token file is in the trained directory