How to extract object nouns from caption? about yolo-world

[wondervictor]

Hi @myhz0606, thanks for your interest in YOLO-World!
Extracting nouns is easy in our work, we simply adopt the NLTK package to do it as follows:

import string
import nltk
from nltk import word_tokenize, pos_tag


nltk.download('punkt')
nltk.download('averaged_perceptron_tagger')


def extract_noun_phrases(text):
    
    tokens = word_tokenize(text)
    tokens = [token for token in tokens if token not in string.punctuation]
    tagged = pos_tag(tokens)
    print(tagged)
    grammar = 'NP: {<DT>?<JJ.*>*<NN.*>+}'
    cp = nltk.RegexpParser(grammar)
    result = cp.parse(tagged)
    
    noun_phrases = []
    for subtree in result.subtrees():
        if subtree.label() == 'NP':
            noun_phrases.append(' '.join(t[0] for t in subtree.leaves()))
    
    return noun_phrases

[wondervictor]

Hi @lxfater, generally, this will not have an impact, unless there are some similar categories or hard-to-distinguish categories among the 5000 words.

[lxfater]

If I have to input a large number of detection categories (1000), will this affect performance? @wondervictor

[wondervictor]

Hi @lxfater, it will not affect and YOLO-World is born for custom vocabularies without re-training.

[lxfater]

If I feed 5,000 words into the model as categories and ask it to recognize any image, does that cause a performance problem compared to feed the categories that only appear in the image? I'm a novice in this field @wondervictor

[wondervictor]

Thanks for your interest. If you have any questions about YOLO-World in the future, you're welcome to open a new issue.

[tuotalimiaomiao]

Can yolo world train its own dataset from scratch in the industrial domain?

[wondervictor]

Hi @tuotalimiaomiao, sure. But I strongly suggest you use the pre-trained weights to initialize, which is very useful.