mirror of
https://github.com/deepseek-ai/DeepSeek-VL2.git
synced 2026-11-05 01:07:23 +00:00
[GH-ISSUE #49] 请问visual grounding测评的prompt和评测代码 #37
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepSeek-VL2#37
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Mr-Bigworth on GitHub (Feb 2, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-VL2/issues/49
您好，我在复现论文表5结果时发现离论文汇报精度存在差距
数据格式为
conversation =[
"role": "<|User|>"
"content":"\n<|ref|>{}<|/ref|>.".format(text_query)"images":["{}".format(image_path)],
}，
{"role":"<lAssistant|>","content":""}
其中text_query为refcoco的文本expression，image_path对应输入图像，想请教能对应上论文评测结果的输入模版应该是什么呢
@CS-FantaC commented on GitHub (Feb 24, 2025):
您好，请问哪些7B以内（最好3B左右的）开源VLM的grounding能力比较强？
@Mr-Bigworth commented on GitHub (Mar 5, 2025):
Qwen2.5-VL效果比较好
@updateforever commented on GitHub (Jul 28, 2025):
您好，请问您复现出来的跟论文报告的差几个点呀？我在自己的数据集上测grounding表现，可能是场景比较复杂deepseekvl只有0.00几的mean iou得分。但qwen2.5vl全系列都在四十多得分左右。我跟你的prompt做的一样，我看官方issue也是推荐这么写prompt的呀。