Incorrect Class and Concept list for imagenet

[raeudigerRaeffi]

Hi,

Thanks for open sourcing the code for your paper. I am working on a paper to mitigate Information Leakage and want to benchmark it using your concept annotated datasets. However I noticed that the files imagenet_per_class and imagenet_classes only contain 998 unique classes instead of the 1000 from imagenet. I was sofar unable to verify whether this extends to the provided concepts annotations aswell. To reproduce my findings you can run

`import json

with open(f"VLG-CBM/concept_files/imagenet_per_class.json") as f:
class_concept_data = json.load(f)

with open(f"VLG-CBM/concept_files/imagenet_classes.txt") as f:
classes = (f.read()).split('\n')

print(len(classes))
print(len(set(classes)))
print(len(class_concept_data))
test_counter = {}
for i in classes:
if i not in test_counter:
test_counter[i] = 0
else:
print(i)
`

Output:
1000
998
998
missile
sunglasses