zhangyuanhan-ai/bamboo

Bamboo: 4 times larger than ImageNet; 2 time larger than Object365; Built by active learning.

bamboo's Issues

How to download full dataset

How to download full bamboo dataset?
Multiple applications have been submitted within a year, but no response has been received. The website link is also disappear.

can't find the config file of Bamboo-Det

Hi, I want to try Bamboo-Det, but it seems that detectron2 needs a configuration file (.yaml) as well as a weight file (.pkl). I can only find the .pkl file from the link you provided. Where can I find the .yaml file?

Explore website link failed.

The explore website you posted in the project failed.
https://opengvlab.shlab.org.cn/bamboo/home
Would you update it later? I would like to get sample images of curtain categories in a convenient way.

common concepts

Great work.
Where can I access the 809 common concepts mentioned in the paper? The 'id2name' in the label system has 304048 concepts, while i am more interested in those 809 common concepts. Hope you can help me!

How to acquire Bamboo-DET

Hi, I've read the paper about Bamboo. It is mentioned that Bamboo-CLS contains ~68M images and Bamboo-DET contains ~3M images. Is Bamboo-DET contained in Bamboo-CLS? I've downloaded all images of Bamboo-CLS, so what I need for Bamboo-DET is only the box-level annotation file?

How to map 115,217 english categories into chinese categories?

I think the Bamboo is a great job. And I find the label system of Bamboo in README, also find the classification-demo on https://huggingface.co/spaces/CVPR/Bamboo_ViT-B16_demo, and I think it will be useful for me to build a subclass-chinese-classification system

I have a question, how to map the class-name of 115217-cls-model on huggingface into Bamboo-label-system:

for example,bamboo['id2name_zh'][node_id] can get the chinese-class-name, and I can get thee class-name of the classification by trainid2name.json. There is not a mapping between bamboo and trainid2name

Thanks!

why has no class of 'head','face' in detect labels

hi， why has no class of 'head','face' in detect labels。 i think the two classes are so inportant .

can't find the Supplementary Material

Great work.
However, I can't find the Supplementary material you mentioned in the description of Fig.3. Can you provide me with the link to download it?

On the other hand, I have a question about how you do Data annotation (Sec 4.1.5) for object detection task:
It seems that the annotator just needs to tell the system if the prediction is right or not. Does that mean an image is valid only if all interesting objects are correctly detected?