Cannot unzip vqa_data.zip file

[wenliangdai]

The provided vqa_data.zip file is too large and the unzip failed (as shown in the figure above).

We have tried to fix it by zip -F file.zip --out file-large.zip but it doesn't work.

Could you please help to provide a better way to process data files? Thanks!

[yangapku]

Hi, since the vqa_data.zip is too large, we have alternatively provided chunked parts of vqa_train.tsv and vqa_test.tsv for downloading. You can try the following code snippet to download the data files:

dist_source=https://ofa-beijing.oss-cn-beijing.aliyuncs.com/datasets/vqa_data
mkdir -p dataset/vqa_data/ && cd dataset/vqa_data/
for idx in `seq -w 00 36`; do wget ${dist_source}/vqa_train_${idx}.tsv; done
for idx in `seq -w 00 08`; do wget ${dist_source}/vqa_test_${idx}.tsv; done
wget ${dist_source}/vqa_val.tsv
wget ${dist_source}/trainval_ans2label.pkl
wget ${dist_source}/vqa_data_md5.txt # for checking the integrity

Each chunked part tsv file contains 50K lines (around 3GB). You can check the integrity of each file with md5sum using the downloaded vqa_data_md5.txt. By concatenating the chunked parts, you will get the complete vqa_train.tsv and vqa_test.tsv files.

cat vqa_train_* > vqa_train.tsv
cat vqa_test_* > vqa_test.tsv # you can also check the md5 using vqa_data_md5.txt

If you still have problems in downloading the files, please do not hesitate to ask me.