Skip to content

微调qwen2.5-vl做点检测的grounding,数据集应该是什么形式 #3883

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jjjjjjj2020 opened this issue Apr 15, 2025 · 2 comments

Comments

@jjjjjjj2020
Copy link

请问有人知道如果用swift微调qwen2.5-vl做点检测的grounding,数据集应该怎么做呢,直接参照swift给的数据集格式,把框的坐标改成点的坐标会不会对性能有影响?

@Jintao-Huang
Copy link
Collaborator

框的坐标就是 两个点表示的

@jjjjjjj2020
Copy link
Author

框的坐标就是 两个点表示的

我的数据集把框转变为了中心点的坐标如下所示可以训练,但是推理的结果总是只有2个点的输出,如图所示,请问这是什么原因造成的呢?
{
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": " find edd"
},
{
"role": "assistant",
"content": ""
}
],
"images": [
"/public/huangjiao/data/EDD-Grounding/EDD-grounding/train2017/test-19506.png"
],
"objects": {
"ref": [
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd",
"edd"
],
"bbox": [
[
124,
778
],
[
1249,
23
],
[
60,
661
],
[
1236,
122
],
[
89,
577
],
[
93,
407
],
[
179,
719
],
[
851,
881
],
[
779,
1389
],
[
1495,
1401
],
[
1668,
1073
],
[
193,
334
],
[
23,
560
],
[
891,
18
],
[
1026,
213
],
[
827,
1668
],
[
1706,
1711
],
[
1600,
1168
],
[
466,
683
],
[
872,
1772
],
[
1491,
1464
],
[
60,
1021
],
[
138,
545
],
[
901,
1870
],
[
1663,
823
],
[
62,
819
],
[
1546,
722
],
[
771,
32
],
[
1881,
1862
],
[
831,
764
],
[
753,
1250
],
[
913,
359
],
[
859,
1054
],
[
1299,
439
],
[
778,
545
]
]
}
},

Image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants