hi，about token file problem

[yangkunhub]

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

[yangkunhub]

run CUDA_VISIBLE_DEVICES=0 python pascalvoc_generate_pseudo_label.py with dataset.train_batch_size=8 Tips are as follows：

[yangkunhub]

the execution result about CUDA_VISIBLE_DEVICES=0 python lip_crop.py 。 The PASCAL dataset I used, why is the train/cls_feat/cls_feat_pascalvoc folder empty 。

[yangkunhub]

你好我看了你的论文 你的代码我也去尝试复现了 但是我不知道为什么执行 CUDA_VISIBLE_DEVICES=0 python pascalvoc_crop.py这个命令应该是利用滑动窗口生产裁剪后的图片，但是这些生成后的图片是存放在哪个目录下呢？然后生成train/cls_feat/cls_feat_pascalvoc这个文件下也是为空这个里面应该放的应该是token 文件把 。再执行CUDA_VISIBLE_DEVICES=0 python pascalvoc _generate_pseudo_label.py with dataset.train_batch_size=2 model.decoder.n_things=16 生成伪标签然后就报错了 。报错如下

这个错误应该是找不到cls-feat文件 。谢谢作者

[yangkunhub]

希望能得到作者指点

[yangkunhub]

球球了 这篇论文对我很有帮助， 我想问运行了pascalvoc_crop.py，提示是成功运行了如图

但是cls-feat-Pascalvoc 文件夹里的文件是空的，如果这个文件为空下一步在运行pascalvoc _generate_pseudo_label.py 就会提示找不到cls-feat-Pascalvoc文件夹里的文件 就会运行失败 如图

。 我用的是Pascalvoc数据集因为其它数据集太大了我就没试，因为我只用的voc数据集所以root/weight/trained里只放了pascalvoc_decoder.pt。root/weight/dino里都正常放的dino_deitsmall8_pretrain.pth这个文件

[zyy-cn]

@yangkunhub hi，您好，抱歉回复迟了。感谢对我们工作的关注。我这边没有复现出这个问题，请问您运行的环境是什么？目录是否具有写权限？

为了排查问题，建议测试如下代码是否能在'train/cls_feat/cls_feat_pascalvoc'文件夹下正常输出'output':

import os
import pickle
import numpy as np

feature_bank_dir = os.path.join(f'train/cls_feat/cls_feat_pascalvoc')
os.makedirs(feature_bank_dir, exist_ok=True)
feat_save_path = os.path.join(feature_bank_dir, 'output')
cls_feat = np.random.random((5, 384))
dictfile = open(feat_save_path, 'wb')
pickle.dump(cls_feat, dictfile)
dictfile.close()

[yangkunhub]

好啦 作者太谢谢你回复我了 问题解决了 因为我在3090上跑的 torch版本有点低了 卸载又换了torch版本 就出现了生成的文件啦

[tanveer6715]

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

Hi I am facing same problem.. My token files and pseudo_label are generated but when I start training on cityscapes dataset it showed me this error. Your help and suggestion will be appreciated.

[zyy-cn]

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

Hi I am facing same problem.. My token files and pseudo_label are generated but when I start training on cityscapes dataset it showed me this error. Your help and suggestion will be appreciated.

Hi @tanveer6715 , I didn't reproduce this problem on my end using the environment described in the README.md file. Could you please provide more detail information about this problem if you trained under the same environment?

[tanveer6715]

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

Hi I am facing same problem.. My token files and pseudo_label are generated but when I start training on cityscapes dataset it showed me this error. Your help and suggestion will be appreciated.

Hi @tanveer6715 , I didn't reproduce this problem on my end using the environment described in the README.md file. Could you please provide more detail information about this problem if you trained under the same environment?

Thanks for your kind response. Actually I am trying it on cityscapes dataset and the training starts well. It seems to be a kind of warning message when I start the training.