May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? #74
Comments
|
Our current approach is to split the multi-label into multiple binary labels, and randomly select 5 labels as supervisory signals during training. You can follow our data processing method for training. |
Hello，I‘m using my own dataset now,but the category of each data may be different,so what should I do? |
|
你可以更清楚的描述这个问题嘛？我不太理解你的意思。 |
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？ |
|
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
…---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
我还有一个问题，就是在训练的时候mask_num设置为5，是不是代表每次就只训练这5个类别，如果我所有的类别都想要训练，应该怎么设置
发自我的iPhone
…------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:18 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
如果你的类别是10，那你可以设置mask num为10，即使有些图片类别没有达到10，他也会从现有类别中重复选择。 另一种思路，你设置为一个较小的数，你在多次迭代过程中总能学习到这些掩码。
…---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:22 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
我还有一个问题，就是在训练的时候mask_num设置为5，是不是代表每次就只训练这5个类别，如果我所有的类别都想要训练，应该怎么设置
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:18 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
那为什么要这样设计随机选择mask，而不和TestingDataset一样，读取图像对应的所有mask
发自我的iPhone
…------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:25 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
如果你的类别是10，那你可以设置mask num为10，即使有些图片类别没有达到10，他也会从现有类别中重复选择。 另一种思路，你设置为一个较小的数，你在多次迭代过程中总能学习到这些掩码。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:22 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
我还有一个问题，就是在训练的时候mask_num设置为5，是不是代表每次就只训练这5个类别，如果我所有的类别都想要训练，应该怎么设置
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:18 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
保持训练的一致性，训练是要进行多轮的，而且每个数据集的类别数不一样。测试只需要对所有类别测一次。
…---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:31 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
那为什么要这样设计随机选择mask，而不和TestingDataset一样，读取图像对应的所有mask
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:25 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
如果你的类别是10，那你可以设置mask num为10，即使有些图片类别没有达到10，他也会从现有类别中重复选择。 另一种思路，你设置为一个较小的数，你在多次迭代过程中总能学习到这些掩码。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:22 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
我还有一个问题，就是在训练的时候mask_num设置为5，是不是代表每次就只训练这5个类别，如果我所有的类别都想要训练，应该怎么设置
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:18 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
|
我明白了，非常感谢你的耐心解答
发自我的iPhone
…------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:34 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
保持训练的一致性，训练是要进行多轮的，而且每个数据集的类别数不一样。测试只需要对所有类别测一次。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:31 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
那为什么要这样设计随机选择mask，而不和TestingDataset一样，读取图像对应的所有mask
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:25 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
如果你的类别是10，那你可以设置mask num为10，即使有些图片类别没有达到10，他也会从现有类别中重复选择。 另一种思路，你设置为一个较小的数，你在多次迭代过程中总能学习到这些掩码。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:22 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
我还有一个问题，就是在训练的时候mask_num设置为5，是不是代表每次就只训练这5个类别，如果我所有的类别都想要训练，应该怎么设置
发自我的iPhone
------------------ Original ------------------
From: Junlong ***@***.***>
Date: Sun,Jan 12,2025 7:18 PM
To: OpenGVLab/SAM-Med2D ***@***.***>
Cc: Yin Chen ***@***.***>, Comment ***@***.***>
Subject: Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multipletags on a single image? How should I set it up during training? (Issue #74)
是这样的，把所有的标签全部变为二值的。没有的就不用管。另一种在线的方案是在现有的标签中随机选择几个mask，然后处理成2值的用于训练。
---- Replied Message ----
| From | Yin ***@***.***> |
| Date | 01/12/2025 19:14 |
| To | OpenGVLab/SAM-Med2D ***@***.***> |
| Cc | Junlong ***@***.***>,
Comment ***@***.***> |
| Subject | Re: [OpenGVLab/SAM-Med2D] May I ask if it supports labeling multiple tags on a single image? How should I set it up during training? (Issue #74) |
你可以更清楚的描述这个问题嘛？我不太理解你的意思。
就是假设我的数据集一共有十个类别，但是并不是每一张图像都包含这十个类别，可能少于十，那么对于没有的类别，对应的mask应该怎么处理呢？是处理成全黑的吗？
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you commented.Message ID: ***@***.***>
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
No description provided.
The text was updated successfully, but these errors were encountered: