imperialnlp / vtlm Goto Github PK

The multi30k data needs to be prepared for MT fine tuning.

parser.add_argument is error. So feature-extractor.py can't use to extractor the photo feature.

In ParallelDatasetWithRegions, indexing with good indices does no work

Hi! When I downloaded these three websites, they all showed that I could not access them. Is there any other way to get these data? Thank you!

CC_URL="https://zenodo.org/record/4646961/files/conceptual_captions_en_de.tar.bz2"
M30K_FEATS_URL="https://zenodo.org/record/4646961/files/multi30k_oidv4_features.tar.xz"
TF_OBJ_URL="https://zenodo.org/record/4646961/files/faster_rcnn_inception_resnet_v2_atrous_oid_v4_2018_12_12.tar.bz2"

The pretrained model should be finetuned on Multi30k dataset.

Congratulations on the interesting work. I am wondering if you can upload the datasets and checkpoints to google drive. The download speed from zenodo.org is extremely slow.

KL divergence should be implemented in the code for the image mask prediction.

Hello, I have a problem that why my process image file(.pkl) size is 13MB, whereas the image feature from the project downloading is only 271KB. And the time I process one picture about 15 seconds, is that normal? Or my parameter is not right?

This is the first time I run this code. When I run the bash train.sh command, I get an error. There is no train.order file. I run the scripts/01-process-multi30k.sh test_2016_flickr script according to the README file to process the data.
But after the command is executed, the train.order is not produced, can you tell me how to get the train.order file, thank you very much, looking forward to your reply