Jie Chen

Results 8 comments of Jie Chen

> 我还复制了 OP 的调查结果。我的步骤: > > 1. 从 OneDrive 位置 ,我下载了文件 和 ,并分别将它们解压缩到 和 。每组包含 1000 张图像。真实的是 JPEG 格式,合成的则是 PNG 格式。`dire/test/lsun_bedroom/lsun_bedroom``adm.tar.gz``real.tar.gz``lsun_test/1_fake``lsun_test/0_real` > 2. 然后,我从 OneDrive 下载了模型`checkpoints/lsun_adm.pth` > 3. 然后,我跑到下载的模型和文件夹`test.py``lsun_test` >...

![Image](https://github.com/user-attachments/assets/1681e14c-0afc-471a-90bb-b8cd4e740f59) Excuse me!How to understand the three sets of training test and verification sets in the data set?

> > [@lukovnikov](https://github.com/lukovnikov) > > 我为 CelebAHQ 图像运行了 compute_dire - 包括 Recons 和 DIRE,奇怪的是,**celebahqsdv2 ckpt 将所有这些真实图像归类为假图像!** > > lsun-adm、lsun_iddpm 和 lsun_stylegan 可以正确分类真实图像,但不能正确分类其他三个 CKPT。非常奇怪,因为 celebahq 检查点是在 celebahq 真实数据集上训练的,我猜? > > **您必须使用可怕的版本而不是原始/原始图像运行。有关更多信息,请参阅...

> > [@lukovnikov](https://github.com/lukovnikov) > > 我为 CelebAHQ 图像运行了 compute_dire - 包括 Recons 和 DIRE,奇怪的是,**celebahqsdv2 ckpt 将所有这些真实图像归类为假图像!** > > lsun-adm、lsun_iddpm 和 lsun_stylegan 可以正确分类真实图像,但不能正确分类其他三个 CKPT。非常奇怪,因为 celebahq 检查点是在 celebahq 真实数据集上训练的,我猜? > > **您必须使用可怕的版本而不是原始/原始图像运行。有关更多信息,请参阅...

![Image](https://github.com/user-attachments/assets/d3eb7298-0271-43cc-bac8-4a9880210d18) After reading your content, is there a problem with setting the dataset like this? The result is 1 in most cases

![Image](https://github.com/user-attachments/assets/3e131c4a-631d-4d69-862c-30559de1f872) How to understand the three sets of training test and verification sets in the data set?

![Image](https://github.com/user-attachments/assets/17eac520-aab9-40ec-977f-2b380154da22) ![Image](https://github.com/user-attachments/assets/6794d988-c03c-4306-8c41-dce835f881ea) I directly run test.py with the test set in dire, and the result is very poor. Then I delete the forged part. Is that what happened?

您好,我对RECEE的实现也比较感兴趣,想获取您处理后的数据集进行尝试,我的工作邮箱是 [email protected],感谢您的帮助。