esg_reports_human_labeled_v2: discrepancy between test corpus size and the number of uploaded images

[sahel-sh]

Hi,
I have downloaded the visdoc-task from hf and extracted the data and images folders.
For all other vidore tasks that I tested the corpus_size under ./data == the number of images under the ./images
However, for esg_reports_human_labeled_v2 there are 230 images under the images folder, while the two test_*.parquet files under data have 1538 rows combined which matches the number of corpus rows in https://huggingface.co/datasets/vidore/esg_reports_human_labeled_v2.
I was wondering if some of the images are missing?

Context:
I need the jsonl corpus for the collection iterator in Pyserini and parquet image bytes are not json serializable.
I save the images locally and keep the path in the jsonl corpus.
I am cross referencing the images from the dataset with those under ./images just to be sure.

[sahel-sh]

I tried merging the three visdoc-task tar files as an alternative and extracting the merged file.
Got the following error:

...
visdoc-tasks/esg_reports_human_labeled_v2/528.png

gzip: stdin: unexpected end of file
tar: Unexpected EOF in archive
tar: Unexpected EOF in archive
tar: Error is not recoverable: exiting now

again while extracting images for esg_reports_human_labeled_v2 and again only 230 were successfully extracted.

s8sharif@orca:/mnt/users/s8sharif/MMEB-V2/visdoc-tasks$ ls visdoc-tasks/esg_reports_human_labeled_v2/*.png | wc -l
230