Open respectbraca opened 1 year ago
您好想请教您一个问题,我在USSS部分,用的是自己的数据220220大小的一个瓦片数据,但是在预训练模型和结构之间出现不匹配现象不知道为什么,您能帮忙回答一下吗? Hello, I would like to ask you a question. In the USSS part, I used my own tile data of 220220 size, but there was a mismatch between the pre-trained model and the structure. I don’t know why. Can you help answer it? For a moment?
你好,你的代码wsss.py吗,还是你说的是demo-wss.py.还有这个训练部分在哪呢
我对你的工作很感兴趣,在对弱监督部分进行复现的时候发现Recall Rate、KC、F1、IOU等指标结果与原文相差过大,我不知道是哪一部分出了问题。值得注意的是,基于我的GPU我将训练时候的生成器和鉴别器batch size分别设置未30和8。我希望你能指导一下。 I am very interested in your work, while reproducing the weakly supervised part I found that the results of Recall Rate, KC, F1, IOU and other metrics are too different from the original text, I don't know which part of it is wrong. It is worth noting that based on my GPU I have set the generator and discriminator batch sizes during training to 30 and 8 respectively. i hope you can guide me on this.
您好,请教您一下,在弱监督部分,作者使用的数据组织形式是和无监督那一块一样的吗?您的训练数据组织形式可以分享一下吗?
你好,你的代码wsss.py吗,还是你说的是demo-wss.py.还有这个训练部分在哪呢
就是demo-wss.py,它也是训练部分
我对你的工作很感兴趣,在对弱监督部分进行复现的时候发现Recall Rate、KC、F1、IOU等指标结果与原文相差过大,我不知道是哪一部分出了问题。值得注意的是,基于我的GPU我将训练时候的生成器和鉴别器batch size分别设置未30和8。我希望你能指导一下。 I am very interested in your work, while reproducing the weakly supervised part I found that the results of Recall Rate, KC, F1, IOU and other metrics are too different from the original text, I don't know which part of it is wrong. It is worth noting that based on my GPU I have set the generator and discriminator batch sizes during training to 30 and 8 respectively. i hope you can guide me on this.
您好,请教您一下,在弱监督部分,作者使用的数据组织形式是和无监督那一块一样的吗?您的训练数据组织形式可以分享一下吗?
应该是不一样的,数据集格式应该不一样,我的是直接下载的CDD官方数据集就可以用。
我对你的工作很感兴趣,在对弱监督部分进行复现的时候发现Recall Rate、KC、F1、IOU等指标结果与原文相差过大,我不知道是哪一部分出了问题。值得注意的是,基于我的GPU我将训练时候的生成器和鉴别器batch size分别设置未30和8。我希望你能指导一下。 I am very interested in your work, while reproducing the weakly supervised part I found that the results of Recall Rate, KC, F1, IOU and other metrics are too different from the original text, I don't know which part of it is wrong. It is worth noting that based on my GPU I have set the generator and discriminator batch sizes during training to 30 and 8 respectively. i hope you can guide me on this.
您好,请教您一下,在弱监督部分,作者使用的数据组织形式是和无监督那一块一样的吗?您的训练数据组织形式可以分享一下吗?
应该是不一样的,数据集格式应该不一样,我的是直接下载的CDD官方数据集就可以用。
收到,谢谢您的回复和解惑
I am interested in your work, what is the organization of the data used in the weak supervision section? Can you share it? 请问弱监督的训练数据集文件目录是什么样的呀?是如何组织的?能分析一下数据集吗?
您好,请教您一下,在弱监督部分,作者使用的数据组织形式是和无监督那一块一样的吗?您的训练数据组织形式可以分享一下吗?
您好,您的训练数据组织形式可以分享一下吗?我使用下载但是无法训练,能指导一下吗?
弱监督数据集用的和无监督数据集是不一样的,无监督数据集可以直接在整张影像上处理,弱监督的话,需要把大图像切分成小图像来读取。
你可以使用BuildingProcess.py这个函数来处理WHU Building数据集,生成适用于弱监督处理的数据集
你可以使用BuildingProcess.py这个函数来处理WHU Building数据集,生成适用于弱监督处理的数据集
BuildingProcess.py在处理时也会得到对应的图像级别标签是吗?感谢你的回复
我对你的工作很感兴趣,在对弱监督部分进行复现的时候发现Recall Rate、KC、F1、IOU等指标结果与原文相差过大,我不知道是哪一部分出了问题。值得注意的是,基于我的GPU我将训练时候的生成器和鉴别器batch size分别设置未30和8。我希望你能指导一下。
I am very interested in your work, while reproducing the weakly supervised part I found that the results of Recall Rate, KC, F1, IOU and other metrics are too different from the original text, I don't know which part of it is wrong. It is worth noting that based on my GPU I have set the generator and discriminator batch sizes during training to 30 and 8 respectively. i hope you can guide me on this.