jwehrmann / retrieval.pytorch Goto Github PK

Hi jwehrmann,
I have followed the README.md to reproduce the experiments.
However, the caption data "train_caps.en.txt" could not be found in the SCAN provided dataset.
There is just "train_caps.txt". Does you miss it, or Should I rename the file?
Thank you very much.

Hi jwehrmann,
I tested the ADAPT-Ens on Flickr30k test set., and got worse results with models from your codes: {"i2t_r1": 0.0, "i2t_r5": 0.1, "i2t_r10": 0.3, "i2t_medr": 1289.0, "i2t_meanr": 1270.931, "t2i_r1": 20.74, "t2i_r5": 27.6, "t2i_r10": 28.94, "t2i_medr": 386.0, "t2i_meanr": 380.0868, "rsum": 77.68}.
I got the Flickr30k datasets from https://iudata.blob.core.windows.net/scan/data.zip What's the reason?

Could not resolve host: scanproject.blob.core.windows.net.
Could anyone provide another dataset link?
Thanks

Hi jwehrmann,
I wonder that how do you get the raw region information(x, y, w, h). Did you run Faster rcnn by yourself or get it from other sources? How to align the region information with the precomputed features from SCAN?
Thank you very much.

Hi jwehrmann,
When I tried to download Pre-trained models, I found that “COCO ADAPT-t2i” and “COCO ADAPT-i2t” can not download, is that my problem?
Thank you very much.

Hi Jonatas,

Great work!
I am wondering, beside multi-lingual, what's main difference or improvement over your ICCV2019 work?

Do they have similar performance?

I can't download data from the link you provided. Could you please offer another way to download the dataset?