BerVAE

This repo holds the Pytorch implementation of BerVAE. Our implementation is based on BTH.

Environment

Python 3.7.0

Requirement

scipy==1.6.3
h5py==3.1.0
torch==1.8.1+cu111
matplotlib==3.4.2
pandas==1.2.4
numpy==1.19.5
tensorboardX==2.5.1

Quick Start

Download Features

VGG features of FCV are kindly uploaded by the authors of SSVH. You can download them from Baiduyun disk.

Please set the data_root and home_root in args.py in both ./utils/ and ./model/. You can place these features to in data_root.

Training BerVAE

After correctly setting the path, you can run train.py to train the model. Models will be saved in ./models.

Testing BerVAE

When training is done, you can run eval.py to test it. mAP files will be save in ./results. The scatter and box-plots of the AP$@K$ values of the query video clips with respect to the uncertainty level of corresponding hash-codes of the queries will also be saved in ./ To evaluate the uncertainty quantification performance using our proposed IDU, you can run eval_uq.py.