Comments (7)
wondervictor
commented on June 3, 2024
1
@bobo0810 您好，fine-tuning会影响pre-train的zero-shot性能（包括泛化性），但不代表fine-tuning完后模型不再具备zero-shot能力。可以根据fine-tuning的数据量和fine-tuning的迭代数可以控制这个影响，如果可以的话，我比较推荐尝试fine-tuning后面head部分的layers，或者尝试引入LoRA。这一块我们目前还没有做很多的探索，后续我们也会尝试用LoRA做fine-tuning，如果有新的结论，我们会及时更新本repo，在此感谢您的关注！
from yolo-world.
wondervictor
commented on June 3, 2024
您好，非常感谢您对YOLO-World的关注，我们目前已经公开了代码和预训练的权重，对于微调，可以加载我们预训练模型来作为初始化权重，并使用较小的learning rate训练(推荐2e-4)，我们将在近期(两天内)公开在COCO/LVIS微调的配置文件，以及custom数据集的构建方式，到时候可以作为参考，欢迎持续关注！
from yolo-world.
qpfhuan
commented on June 3, 2024
您好，非常感谢您对YOLO-World的关注，我们目前已经公开了代码和预训练的权重，对于微调，可以加载我们预训练模型来作为初始化权重，并使用较小的learning rate训练(推荐2e-4)，我们将在近期(两天内)公开在COCO/LVIS微调的配置文件，以及custom数据集的构建方式，到时候可以作为参考，欢迎持续关注！
太牛了！！！请问，除了微调训练，在text prompt上有什么建议吗？
from yolo-world.
xiyangyang99
commented on June 3, 2024
您好，非常感谢您对YOLO-World的关注，我们目前已经公开了代码和预训练的权重，对于微调，可以加载我们预训练模型来作为初始化权重，并使用较小的learning rate训练(推荐2e-4)，我们将在近期(两天内)公开在COCO/LVIS微调的配置文件，以及custom数据集的构建方式，到时候可以作为参考，欢迎持续关注！
作者你好，你们的离线词汇的方式是什么啊？是文本和图像分开的吗？groundingdino因为交叉注意力的计算不得不同时进行，yolo-world好像感觉是推理的时候直接将训练好之后的词汇权重或者参数直接拿过来用，是clip的那种方式吗？
from yolo-world.
wondervictor
commented on June 3, 2024
@qpfhuan 您好，我们目前提供了一些关于fine-tuning的细节以及相应的config，在COCO和LVIS的权重我们也将快速上传，您可以参考 docs/fine-tuning这个介绍在自己的数据集上尝试一下，如果有不清楚的地方，欢迎讨论！
from yolo-world.
bobo0810
commented on June 3, 2024
@wondervictor 您好，麻烦问下 基于下游的闭集检测数据微调后是否还保留原来的泛化性呢？
from yolo-world.
wondervictor
commented on June 3, 2024
Thanks for your interest. If you have any questions about YOLO-World in the future, you're welcome to open a new issue.
from yolo-world.
Related Issues (20)
- How to make predictions using image bytes instead of image paths HOT 5
- Normal for confidence level to always be at 1.00? HOT 2
- ModuleNotFoundError: No module named 'mmcv._ext' HOT 2
- Do you have a more detailed network model structure for YOLO World? HOT 2
- FileNotFoundError: [Errno 2] No such file or directory: 'configs/pretrain/yolo_world_v2_x_vlpan_bn_2e-3_100e_4x8gpus_obj365v1_goldg_train_1280ft_lvis_minival.py' HOT 3
- YOLO WORLD for multilabel task HOT 1
- OSError: Incorrect path_or_model_id: '../pretrained_models/clip-vit-base-patch32-projection'. Please provide either the path to a local folder or the repo_id of a model on the Hub. HOT 3
- 为什么我用自己制作的coco数据集微调模型后，训练结果中的bbox_mAP_s一直为-1.000，其他数据都正常 HOT 6
- Hoping for some guidance in open set detection finetuning (Long post) HOT 1
- why?texts = [[t.strip()] for t in args.text.split(',')] + [[' ']] HOT 3
- onnx模型导出成功，但是运行推理失败。
- Training visualization
- installation is crazy ! HOT 7
- Region-Text Matching question
- 哪位大神帮我看看 按照大家的步骤来的 我真的不知道哪里出错了 。。。。。。。 HOT 2
- 小tips HOT 1
- 测试结果惨不忍睹 HOT 17
- Thanks for your work,excellent! some question about yolo-world finetune freeze, glip Pseudo label and prompt.
- the question about image_demo HOT 2
- I would like to ask how it uses YOLOv8 as the backbone in his code, and specifically in which Python file the forward computation for the backbone is performed? HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from yolo-world.