microsoft/X-Decoder

A little bug in `vlpencoder.py`

geweihgg opened this issue · 1 comments

geweihgg commented

code:

X-Decoder/xdecoder/language/vlpencoder.py

Line 119 in a85d2f4

clss_embeddings.append(extract_mean_emb(class_names))

I think this should be:

if prompt:
    for clss in class_names:
        txts = [template.format(clss.replace('-other','').replace('-merged','').replace('-stuff','')) for template in templates]
        clss_embeddings.append(extract_mean_emb(txts))
else:
    for clss in class_names:
        clss_embeddings.append(extract_mean_emb([clss]))
MaureenZOU commented

Thanks for your careful! Just fix this bug : )