Skip to content

Latest commit

 

History

History
43 lines (38 loc) · 2.52 KB

data.md

File metadata and controls

43 lines (38 loc) · 2.52 KB
Raw
Data Description Download
Visual Genome ln -s VG/images data/vg/images Official
MSCOCO 2014 ln -s coco2014/images data/refcoco/images Official
Converted annotations unzip data.zip OneDrive
Meteor package unzip meteor.zip -d controlcap/common/evaluation/ OneDrive
Pre-trained ControlCap weights and logs (Optional) mv <your_path>/ckpts/* ckpts/ OneDrive

P.S. Files in BaiduDrive, the passpord is (3g1k).

To train and evaluate ControlCap, download the files in the table and arrange the files according to the file tree below. (Uploading)

    |--ControlCap/
      |--data/
        |--vg/
           |--controlcap/
           |--images/
              |--1000.jpg
              |--1001.jpg
              ...
        |--refcoco
           |--controlcap/
           |--images/
              |--COCO_train2014_000000000009.jpg
              |--COCO_train2014_000000000025.jpg
              ...
           ...
      |--ckpts/
         |--vg1.2_refcocog_5e.pth
         |--refcocog_gt.pth
      |--configs/
      |--controlcap/
      |--docs/
      |--scripts/
      |--train.py
      |--eval.py

P.S. The converted annotations are generated using data.sh, the original annotations are as follows:

  1. annotations of Visual Genome for dense captioning.

  2. test_caption.json and mdetr_annotations of GlaMM for evaluating referring expression generation.