About Multi-label Training Scheme about imagenet21k HOT 15 CLOSED

Could you provide the code of how to get the multi-label of each image according to words.txt file

all the transformation and semantic data can be found in the resource file. make sure you review it thoroughly
https://miil-public-eu.oss-eu-central-1.aliyuncs.com/model-zoo/ImageNet_21K_P/resources/winter21/imagenet21k_miil_tree.pth

you are also welcome to validate them via the link @cissoidx sent (that was my source for generating the tree)

from imagenet21k.

review thoroughly https://github.com/Alibaba-MIIL/ImageNet21K/blob/main/train_semantic_softmax.py file.

specifically the conversion is done in

from imagenet21k.

Does the multi-label here refer to the multi-label of the same kind of the sematic tree, such as animal-dog-border_collie? But if a picture contains multiple different types of labels, such as animal-dog-border_collie, human-woman-young_woman, how to solve this problem?

from imagenet21k.

So you extract the meaning of each category in imagenet22k, and then use wordnet to construct a semantic tree. Finally, the semantic tree is used to generate multi-label information for each category, that is, the multi-label information of all images in each directory of imagenet22k is the same.

from imagenet21k.

So you extract the meaning of each category in imagenet22k, and then use wordnet to construct a semantic tree. Finally, the semantic tree is used to generate multi-label information for each category, that is, the multi-label information of all images in each directory of imagenet22k is the same.

i did not understand if there is a question or not. your description seems correct

from imagenet21k.

Does the multi-label here refer to the multi-label of the same kind of the sematic tree, such as animal-dog-border_collie? But if a picture contains multiple different types of labels, such as animal-dog-border_collie, human-woman-young_woman, how to solve this problem?

it won't. the base tagging of original imagenet21K is single label.

from imagenet21k.

Where can I find the mapping between class ids and class names of the original ImageNet-21K?

from imagenet21k.

https://raw.githubusercontent.com/niharikajainn/imagenet-ancestors-descendants/master/words.txt

from imagenet21k.

Thank you very much

from imagenet21k.

Could you provide the code of how to get the multi-label of each image according to words.txt file

from imagenet21k.

Could you provide the code of how to get the multi-label of each image according to words.txt file

same question here

from imagenet21k.

Could you provide the code of how to get the multi-label of each image according to words.txt file

@SJLeo

https://github.com/niharikajainn/imagenet-ancestors-descendants
check the categories.py file

from imagenet21k.

Could you provide the code of how to generating the tree

from imagenet21k.

No at the moment. you are welcome to write a code of your own and validate the tree.

from imagenet21k.

Thank you very much

from imagenet21k.