About the dataset VQA2.0

[alice-cool]

1. could you tell me the tsv file that is noted by red rectangles in the below picture whether it comes from peterson bottom up and top down github(https://github.com/peteanderson80/bottom-up-attention). I use tsv file from peterson and for example use the convert_test_lmdb.py file to run and produce the lmdb file. The same is for trainval2014(10100)， and test2015(10100) But I run the eval_tasks.py
2. When I RUN the eval_tasks.py
   . python eval_tasks.py --bert_model bert-base-uncased --from_pretrained save/VQA_bert_base_6layer_6conect-pretrained/pytorch_model_19.bin --config_file config/bert_base_6layer_6conect.json --task 0 --split minval
   the result is so bad . The score is zero .
3. if the dataset doesn't come from peterson 10~100 features from faster rcnn . Could you share the tsv dataset. Because dropbox sharing link is so unstable, always end abruptly...